Plastic Recycling Innovation: Evidence from Patent Portfolios and Convergence

Abstract

Plastic recycling technologies are rapidly being reoriented toward process, operations, and quality-centered innovation, driven by the circular economy and digital transformation. Using 64,639 triadic patents (2005–2024), this study applies International Patent Classification (IPC) portfolio, co-occurrence network, and BERTopic analyses to compare technological structures before and after 2015. Since 2015, data- and AI-enabled sorting and process optimization (IPC class G06), tracking and connectivity (IPC class H04), collection and logistics (IPC class B65), water treatment (IPC class C02), and quality modification/compounding (IPC class C09) have expanded, while organic chemistry (IPC class C07), signal-processing circuitry (IPC class H03), and petroleum/fuel conversion (IPC class C10) have declined. G06 and H04 together account for approximately 29% of the total portfolio and record the largest share increases (+1.63 and +1.28 percentage points); water treatment (C02F) and quality correction (C09K) expand by 0.62 and 0.38 percentage points, while organic chemistry (C07) shows the largest decline (−2.16 percentage points). Topic modeling identifies 10 topics in 2005–2014 and 11 in 2015–2024, with the later period newly featuring reverse logistics for reusable packaging, remanufacturing, chemical recycling for packaging, and data sources. Cross-domain network linkages rise from 49 to 68, with processing–logistics and post-treatment–standardization combinations showing the strongest structural strengthening. Industrially, these findings offer reference signals for firms aligning R&D and IP portfolios with domains of concentrated innovation, particularly AI-enabled sorting, digital connectivity, and feedstock quality correction. For policy, the strengthening of cross-domain linkages suggests that support for sorting infrastructure, traceability and data standards, and quality certification frameworks targets where R&D effort is most concentrated.

1. Introduction

Plastic pollution has become a defining global environmental challenge, with a substantial share of plastics produced worldwide ending up landfilled or accumulating in ecosystems rather than being recycled [1,2]. The slow degradation of plastics and their transformation into microplastics amplify long-term hazard burdens [3,4]. The material’s carbon intensity further makes recycling a key lever for circular economy and decarbonization goals [5]. Mechanical recycling alone remains limited by contaminated feedstocks and property degradation [6]. Consequently, the field is increasingly viewed as requiring system-level transformation across the full value chain from collection and sorting to washing, quality control, and reintegration [7,8,9]. Industry 4.0 technologies, including sensors, the Internet of Things (IoT), and Artificial Intelligence (AI) analytics, have accelerated this transition by enabling process optimization, automated sorting, and traceability across heterogeneous waste streams [10,11,12]. EU policy has reinforced this trajectory: the 2015 and 2020 Circular Economy Action Plans, the Single-Use Plastics Directive, and the Eco-design for Sustainable Products Regulation collectively provided strong signals directing R&D toward circular system implementation rather than individual unit processes [13,14]. However, how these structural shifts are reflected in the joint evolution of patent portfolios, convergence networks, and semantic topics around the 2015 inflection point remains unexamined at scale.

Many existing studies on research trends in plastic recycling rely on bibliometric analyses of academic publications. While useful for tracing scientific knowledge accumulation, such analyses are limited in detecting early signals of industrial application and R&D orientation [15]. Patents, by contrast, are outcomes of technological development tightly coupled with firms’ market strategies and have long served as quantitative indicators of innovation direction and pace [16,17]. Triadic (US–Europe–Japan) patents entail substantial international filing costs and therefore represent relatively high-value inventions suitable for monitoring global innovation signals [18]. In the plastics domain, patent-based analyses have demonstrated their practical utility for technology and policy decision-making by synthesizing innovation directions such as recycling, circular design, and alternative feedstocks [7].

Nevertheless, prior patent trend studies often adopt narrow analytical scopes, providing limited integrated explanation of which technology combinations strengthen across the full value chain [19]. From the perspective that technological innovation emerges through the recombination of existing knowledge, combinational analyses based on co-occurring classification codes, and network structures are effective for structurally interpreting the strengthening or weakening of technological convergence [20]. In addition, topics that rise sharply at specific points can be quantitatively captured through burst detection, which helps clarify “when” trend transitions occur within long time-series data [21].

Visualizing changes in knowledge structures from the perspectives of research fronts and intellectual bases can help interpret the context of emerging themes in complex technological domains [22]. Topic modeling has also been widely used to uncover semantic structures in large-scale patent texts; while Latent Dirichlet Allocation (LDA) has been broadly adopted, it has been criticized for insufficiently reflecting contextual information [23]. With the spread of contextual embedding methods, topic modeling that leverages Bidirectional Encoder Representations from Transformers (BERT) embeddings have evolved to improve topic coherence and interpretability [24]. BERT-based Topic Modeling (BERTopic) combines embeddings, clustering, and class-based Term Frequency-Inverse Document Frequency (TF–IDF) to derive semantically grounded topics from large corpora and can be applied to technical documents such as patent abstracts and claims [25].

Despite growing patent-based evidence on individual recycling processes, three specific gaps in the existing literature remain unaddressed. First, prior studies largely focus on single technology domains or rely on keyword-frequency categorization, providing limited insight into which combinations of technologies strengthen across the full recycling value chain [19]. Second, with the exception of a closely related recent study, comparative analysis explicitly structured around a pre/post-2015 inflection point has not been systematically applied to a large-scale triadic patent corpus. Third, no published study has integrated IPC portfolio analysis, IPC co-occurrence network analysis, and BERTopic-based topic modeling within a unified analytical framework applied to plastic recycling innovation.

The present study addresses all three gaps using 64,639 triadic patent families (2005–2024). Jointly analyzing portfolio, convergence, and semantic dimensions around the 2015 inflection point is important because it reveals which technology combinations are being reinforced across the full value chain, when structural transitions occur, and how the innovation agenda has semantically reorganized dimensions that single-method or single-domain studies cannot jointly capture. This integrated evidence is necessary for firms aligning R&D portfolios with emerging convergence zones and for policymakers prioritizing infrastructure and standards in domains where innovation has demonstrably concentrated.

Against this backdrop, this study structures the innovation landscape of plastic recycling technologies from a value-chain perspective using patent data and comprehensively examines how post-2015 changes in technological portfolios, technological convergence, and semantic topic dynamics co-evolved. Comparing the periods 2005–2014 and 2015–2024, we (1) identify changes in IPC-based portfolios at the section, class, and subclass levels; (2) assess the strengthening and weakening of convergence patterns using IPC co-occurrence networks; and (3) extract patterns of topic emergence, disappearance, and transition through topic modeling, thereby identifying emergent technology groups and structurally strengthening convergence combinations as observable in the patent portfolio.

Academically, this study extends plastic recycling research beyond unit-process-centered perspectives by proposing an integrated framework that characterizes structural shifts in the plastic recycling innovation portfolio through the combination of classification-based and semantic topic analysis. Practically, it provides patent-based evidence that may serve as a reference signal for firms aligning R&D and IP portfolios with areas of concentrated innovation activity, and for assessing operational capability gaps in water treatment, quality correction, and logistics. From a policy perspective, it offers evidence that may inform discussions about priorities for infrastructure and standards—such as collection, sorting, quality management, and traceability systems—aligned with the domains where patent activity has demonstrably concentrated since 2015.

2. Related Literature and Hypothesis Development

2.1. Technology and Policy Developments in Waste Plastic Recycling

Research on waste-plastic recycling gained momentum as scholars began to systematically synthesize the potentials and limitations of conversion technologies capable of handling mixed and contaminated waste streams [26]. Subsequent work extended beyond comparisons of single-process performance toward a systems perspective, emphasizing that the overall efficiency and quality of recycled outputs are determined when sorting, pretreatment, and reprocessing are sequentially coupled. In parallel, the discussion has increasingly recognized that mechanical and chemical recycling can assume differentiated roles depending on feedstock composition, contamination levels, and target quality requirements [27,28]. Collectively, this literature highlights that recycling bottlenecks repeatedly converge on feedstock heterogeneity and quality degradation, implying that advances in collection and sorting must accompany process innovation [29].

From an environmental-impact perspective, evidence has accumulated in favor of condition-dependent comparisons rather than definitive claims of superiority across technologies. Life-cycle assessments (LCAs) comparing pyrolysis-based chemical recycling of mixed waste plastics with mechanical recycling and energy recovery show that conclusions can vary substantially with assumptions regarding the electricity mix, product-quality scenarios, and system boundary definitions [30]. A review of LCAs covering both bio-based and fossil-based plastics further notes that sensitivity is strongly driven by functional units, allocation approaches, and data quality [31]. In addition, reports by public agencies have structured key sources of uncertainty in life-cycle impact assessments of plastics-such as boundary choices, emission factors, and scenario assumptions, thereby clarifying critical cautions for comparative interpretation [32].

Quality and hazard concerns are particularly salient in relation to the fate and behavior of additives and contaminants. Studies synthesizing the migration, release, and environmental impacts of plastic additives have been used to explain potential hazards and quality variability in recycling processes [33]. Beyond this, prior work has documented how product design, source separation practices, detection limits in sorting, and property deterioration in recycled materials jointly create trade-offs [34], alongside emerging attempts to quantify the technical upper bounds of packaging circularity [35]. Comparative analyses of packaging recycling systems across European countries, further indicating that differences in institutions, source-separation arrangements, and sorting infrastructure translate directly into recycling quality and costs [36].

More recently, as recognition has grown that high-purity sorting and quality correction are instrumental in the real-world performance of the circular economy, research on sensor, spectroscopy, and AI-enabled automated sorting has intensified. Systematic reviews of sensor technologies for solid-waste sorting compare application conditions as well as strengths and limitations across sensor types [37]. Reviews addressing the integration of IR/Raman/Laser-Induced Breakdown Spectroscopy (LIBS) with chemometrics emphasize that data preprocessing, robustness to contamination and coloration, and model generalization are central challenges for deployment in operating facilities [38]. Hyperspectral-based plastic detection and sorting studies further discuss opportunities to extend these approaches beyond recycling plants to environmental monitoring [39]. In parallel, for PET in particular, advances that substantially improved enzymatic depolymerization performance have attracted attention by demonstrating the feasibility of recovering high-quality feedstocks [40].

On the policy and standards front, a growing body of work argues that technological upgrading becomes more effective when coupled with institutional and information infrastructures. ISO 15270 has been widely used as a representative guideline that outlines quality requirements and options across the full chain of plastics waste recovery and recycling [41]. In the context of transboundary movement, the Basel Convention’s plastic waste amendments have altered rules governing international trade and shipment, directly affecting constraints faced by the recycling industry [42]. Regulations targeting single-use plastics have also been institutionalized in ways that simultaneously drive product redesign, substitution, and collection systems [43]. Moreover, the EU’s Eco-design for Sustainable Products Regulation (ESPR) provides a framework for codifying product-level circularity requirements, supporting the need to integrate improved recyclability with information-based management, including traceability and standardization [44]. At a macro level, circular-economy syntheses and reports from international organizations consistently argue that pollution reduction is difficult to achieve without applying reduction, reuse, substitution, and recycling as an integrated package [45,46,47,48].

2.2. Bibliometric and Patent-Based Research on Waste Plastics

Technological change in the waste-plastics and recycling domain has been quantified by combining publication-based (bibliometric) analyses with patent-based analyses of innovation activity. Patents are widely used as core data for technology landscape studies because they can capture, relatively early, signals of competitors’ R&D directions, the reconfiguration of technological axes, and potential pathways to commercialization [49,50,51]. An Organization for Economic Co-operation and Development (OECD) study proposed a framework that quantifies plastic innovation linked to the environment and circularity through patents and interprets policy and technological change in tandem [50], while a European Patent Office (EPO) Technology Insight report structured plastic waste management technologies over a long-time horizon and highlighted a strengthening trend toward sorting, purification, and digitally enabled management [51].

A closely related recent study also applied patent-based analysis to plastic recycling innovation, with a focus on thematic clustering and temporal trends. The present work differs in three key respects: it employs triadic patent family data within a unified framework combining IPC portfolio analysis, co-occurrence network analysis, and BERTopic-based topic modeling; it explicitly uses 2015 as an empirically motivated inflection point for systematic before and after comparison; and it extends through 2024 with topic-transition mapping that tracks how thematic structures were reorganized across periods. These differences yield complementary and more fine-grained insights into the structural and semantic dimensions of plastic recycling innovation.

In publication-based trend analysis, a growing body of bibliometric research has derived research clusters and identified growth areas using keyword co-occurrence and citation networks [52,53,54]. Visualization of Similarities viewer (VOSviewer 1.6.20) and bibliometrix have been widely adopted as implementation tools, supporting large-scale network visualization and reproducible workflows [55,56]. However, because signals of academic expansion do not necessarily coincide with signals of industrialization, there is an increasing emphasis on cross-validation with, or integrated analysis of, patent-based results.

Methodologies for patent analysis have also evolved beyond simple application-count trends toward network analysis, mapping, and statistical modeling. Patent citation network analysis, for instance, enables the structural interpretation of knowledge diffusion and the distribution of technological influence [57]. Classification (IPC/Cooperative Patent Classification (CPC))-based portfolio mapping and overlay analysis are particularly effective for comparing technology landscapes across firms, countries, and time periods, and “difference map” approaches have been proposed to visually contrast relative strengths and weaknesses among competing actors [58]. Studies using global IPC maps (overlays) have further refined landscape comparisons by visualizing technological distance and positioning [59]. Algorithmic approaches that link IPC to other economic and industrial classifications via concordance have also suggested avenues for extending long-term comparisons and convergence analyses [60]. In addition, research leveraging IPC/CPC classifications to extract applicants’ technological patterns and their temporal dynamics complements trend analysis at the level of innovation actors [61]. Analyses that track dynamic patterns of technological convergence using IPC co-occurrence provide a framework for quantifying changes in the convergence structure itself [62].

In parallel, studies have systematized how patent information can be used from a strategic technology management perspective such as competitive monitoring, technology valuation, and portfolio management thereby clarifying how patent indicators connect to decision-making processes [63]. Approaches that combine bibliometrics and patents have also been proposed to forecast and anticipate emerging technologies [64]. Text-mining-based patent networks have been suggested as tools for trend detection in high-technology fields [65], and SAO-based patent intelligence systems illustrate how semantic structures can be extracted from patent texts to support mapping and network analysis [66]. Keyword-based patent maps have become a representative method for exploring new technology opportunities [67], and two-stage patent analyses that customize opportunity exploration by incorporating firm capabilities have also been proposed [68]. Moreover, patent-based technology opportunity discovery and roadmap construction have expanded to include opportunity identification using text and classification information [69], patent roadmaps for competition and strategy formulation [70], and data-driven roadmaps based on association rules [71]. Approaches that analyze developmental patterns in patent texts to build patent roadmaps are often presented as representative cases.

Finally, interpretation of patent trends becomes more persuasive when combined with up-to-date syntheses of the underlying technologies. Reviews that summarize the status and outlook of chemical recycling provide background for interpreting patent trends by delineating process categories and differences in technological maturity [72,73,74,75]. Likewise, reviews on AI-enabled waste management synthesize trends in which classification, sorting, and logistics optimization are increasingly strengthened through integration with digital technologies, offering a basis for interpreting signals of digital transformation [76,77].

2.3. Methodological Studies Related to IPC

The International Patent Classification (IPC) is an international standard classification system that organizes patents by technological domain, and its structure and rules of use are summarized in World Intellectual Property Organization (WIPO)’s guidance documents [78]. A central issue in long-run time-series analysis is code migration caused by revisions to the classification scheme; WIPO explains the purpose and use of the Revision Concordance List (RCL) in its FAQ materials [79]. The practical basis for revision procedures and consistency management is further detailed in the IPC revision guidelines [80]. The Cooperative Patent Classification (CPC) is used as a more granular system that extends IPC, and CPC–IPC concordance tables are useful for conversion across classification systems and for comparative analyses [81,82]. In addition, documents that map IPC codes to technology fields—thereby enabling cross-country and cross-field comparisons—have served as a de facto standard reference for defining technology fields, improving both analytical boundaries setting and the comparability of results across studies [83]. Accordingly, IPC-based trend analysis should be designed to jointly address (1) revisions and consistency management, (2) code mapping and crosswalk-based conversion, and (3) portfolio- and network-based interpretation, to ensure reproducibility and comparability.

3. Data and Methodology

The purpose of this study is to quantitatively identify the long-term technological axes and structural changes in waste-plastic recycling technologies using patent data from 2005 to 2024, and to derive core technological axes and convergence relationships by comparatively analyzing the reconfiguration of technology portfolios and directions of thematic evolution around the inflection point of 2015. While conventional chemistry- and process-based technologies remain salient, we seek to empirically interpret how the expanding trend of digitally integrated technologies—such as data- and AI-enabled sorting and quality management, tracking and connectivity, and collection and logistics—materializes as changes in both classification-based structures and meaning-based thematic structures.

To this end, we collected and curated 64,639 triadic patent families filed between 2005 and 2024 to construct an analysis corpus, and then split the observation window into two periods (2005–2014 and 2015–2024) to apply an identical analytical workflow to each. The analysis consisted of (1) technology portfolio analysis by computing IPC section-, class-, and subclass-level distributions and change rates; (2) identification of inter-technology coupling structures by constructing IPC co-occurrence networks; and (3) interpretation of thematic structures and transition patterns through period-specific topic modeling and topic–IPC subclass mapping. As illustrated in Figure 1, the research procedure proceeded through data collection, data curation, IPC-based quantitative analysis, and topic-evolution analysis linked to IPC. By cross-examining the outputs generated at each stage, we synthesized the timing, direction, and structural implications of technological transition in a coherent manner.

3.1. Data Collections

To empirically examine long-term changes in technologies related to waste plastic recycling and resource circularity, this study adopts patent data as its primary analytical source. Patents represent codified knowledge outputs whose novelty and industrial applicability have been assessed to a certain extent, making them well suited for long-horizon comparisons and for tracking shifts in both the composition of technological domains and their combinational relationships. Accordingly, we collected triadic patent applications from the Wips On patent database, yielding a final analytical sample of 64,639 patent applications filed over the period 2005 to 2024.

The overall observation window spans 20 years from 2005 to 2024. To account for exogenous changes that could plausibly serve as inflection points in technological development, such as the diffusion of circular economy initiatives, digital transformation, and the broader adoption of data-driven operational technologies, we split the period into two subperiods centered on 2015, namely 2005 to 2014 and 2015 to 2024. This design allows us to move beyond assessing simple growth in patenting activity and instead compare how technological structures and thematic systems were reconfigured across the two periods. In addition, we standardized the rights stage to patent applications to minimize timing distortions arising from grant lags and cross-country differences in examination systems.

As shown in

Table 1. Search Formula.

Search Formula
Database: Wips On patent database, Search period: 2005–2024, Patent type: Triadic patent applications
EN_ALL: (‘plastic waste’ OR ‘waste plastic’ OR ‘polymer waste’ OR ‘post-consumer plastic’ OR ‘post consumer plastic’ OR ‘post-industrial plastic’ OR ‘post industrial plastic’ OR PET OR rPET OR ‘polyethylene terephthalate’ OR PE OR ‘polyethylene’ OR PP OR ‘polypropylene’ OR PS OR ‘polystyrene’ OR PVC OR ‘polyvinyl chloride’) AND EN_ALL: (waste OR scrap OR discarded OR refuse OR ‘post-consumer’ OR ‘post consumer’ OR ‘post-industrial’ OR ‘post industrial’) AND EN_ALL: (recycling OR ‘mechanical recycling’ OR ‘chemical recycling’ OR reuse OR ‘re-use’ OR upcycling OR recovery OR reclaim OR depolymer OR solvolys OR hydrolys OR glycolys OR methanolys OR pyrolys)

, the search query was constructed by combining keywords covering waste plastics and plastic waste and major resins such as Polyethylene Terephthalate (PET), Polyethylene (PE), Polypropylene (PP), Polystyrene (PS), and Polyvinyl Chloride (PVC), terms representing circular activities such as recycling, recovery, reuse, and upcycling, and terms related to chemical and physical treatment processes including depolymerization, solvolysis, and pyrolysis. This design simultaneously reflects three dimensions, namely materials as target resins, activities as circular actions, and processes as treatment mechanisms, thereby defining a search scope that avoids undue bias toward any single process route or polymer type.

3.2. Pre-Processing

The collected raw data were subjected to a curation process to ensure internal validity. First, duplicate patents were removed to prevent the same invention from being counted multiple times. Duplicate records arising from patent-family overlaps or database integration can introduce systematic bias in estimating the scale of technologies by domain; therefore, duplicates were identified and eliminated using patent identifiers and bibliographic fields. Second, documents that could be captured by the search query but were non-plastic in scope or otherwise outside the study boundary were excluded. This step was intended to keep the analytical sample consistently aligned with the thematic scope of waste plastic recycling and resource circularity.

Third, IPC information was collected for both primary and additional assigned codes and then standardized through a normalization procedure to enable aggregation at the section, class, and subclass levels. Specifically, we harmonized discrepancies arising from spacing and delimiter conventions, version-related differences in code notation, and potential inconsistencies introduced when truncating codes to class or subclass levels. This preprocessing was designed to minimize code-matching errors in subsequent portfolio analyses, including share and growth-rate calculations, as well as in IPC co-occurrence network analysis.

3.3. IPC-Based Technology Portfolio Analysis

This section applies IPC-based portfolio analysis at the section, class, and subclass levels to identify which technological domains expanded and how co-occurrence patterns be-tween domains shifted across the two periods.

3.3.1. Section Distribution Analysis

At the IPC section level, we examine the macro-level structure of the target technology set and assess whether the portfolio is concentrated in a single section or distributed across multiple sections. We compute patent counts and shares by section to summarize the overall structure, and we compare the two periods, 2005 to 2014 and 2015 to 2024, to identify the direction of shifts in section composition. This analysis provides a macro-level check on whether chemistry and process-oriented sections remain dominant and whether digitally related sections, including data processing, communications, and measurement, expand in relative terms.

3.3.2. Yearly Trends in IPC Sections

The year-by-year section analysis is conducted to examine the continuity of long-run trends and to identify potential inflection points. For each year, we calculate section-level shares and assess the direction and pace of annual change. This allows us to distinguish sections that expand sharply after a particular period from those that contract gradually, and to further verify how the timing of portfolio reallocation relates to the period around 2015. When necessary, auxiliary indicators such as moving averages can be used to separate short-term fluctuations from long-term trends for interpretation.

3.3.3. Class Analysis

Class-level analysis is conducted to identify the core technological axes of the portfolio. After truncating IPC codes to the class level, we compute patent counts, shares, and rankings for each class. We also examine the cumulative share accounted for by top classes to assess the degree of concentration and dispersion in the technology portfolio. These results provide a basis for discussing categories such as core classes, defined as large-scale domains; growth classes, defined as domains with rapid increases; and niche classes, defined as relatively small but persistent domains.

3.3.4. Class Change-Rate Analysis

To more directly capture technological structural transitions, we compute the difference in class shares between the two periods as a change rate expressed in percentage points. This metric helps distinguish whether growth in absolute counts reflects overall increases in patenting activity or a genuine shift in portfolio composition. By analyzing changes in shares, we identify rapidly rising and rapidly declining classes and quantitatively present technological axes that gained importance after the transition as well as those whose relative importance diminished. For classes with the largest changes, we further interpret the substantive content of the shift by linking them to related subclasses and topic-modeling results.

3.3.5. Subclass Analysis

Subclass-level analysis traces the drivers of change at a more granular technological level. We compute counts, shares, and period-by-period changes for each subclass to assess which specific subdomains underpin expansions or contractions observed at the class level. As different subclasses within the same class can exhibit opposing trends, subclass analysis helps explain the micro-level mechanisms of portfolio reconfiguration. When needed, subclass–keyword linkages can be used to further verify whether classification-based changes align with meaning-based changes derived from text analysis.

3.4. Topic Evolution and IPC Mapping Analysis

This section applies BERTopic-based topic modeling to the patent texts of each period to compare thematic structures and trace topic transitions, and maps the resulting topics to IPC subclasses to examine how meaning-based thematic change aligns with the classification-based portfolio shifts identified in Section 3.3. An identical analytical workflow is applied to both periods to ensure comparability.

3.4.1. BERTopic Modeling

Patent texts in each period are treated as the unit of analysis, and an embedding-based topic model is constructed. We then apply class-based TF–IDF to the clustering results to extract representative keywords for each topic, and we examine the distribution and proximity of topics in the thematic space through distance-based visualization. In addition, hierarchical clustering is used to infer topic hierarchies as a complementary structure, enabling interpretation not only of parallel relationships among similar topics but also of broader inclusion relationships.

3.4.2. Topic Identification

In the topic identification stage, we assign topic labels by reviewing representative keywords and representative documents, or representative sentences, for the topics derived in each period. Labeling is conducted to reflect the technical content of each topic while maintaining consistent naming conventions to support subsequent comparative interpretation. Topics are also organized to align with value-chain stages such as process, operations, quality, traceability, logistics, and conversion, allowing us to capture how themes can differentiate by application context even within the same technological domain.

3.4.3. Comparative Interpretation

Comparative interpretation distinguishes persistent, emerging, and disappearing topics across the two periods and examines topic movements in terms of merging, splitting, and thematic transition. Topics observed in both periods are treated as persistent, topics that rise after 2015 are classified as emerging, and topics that decline in prominence or disappear are categorized as diminishing or disappearing. In addition, the analysis differentiates cases in which similar keyword clusters are consolidated into a single topic, interpreted as merging, from cases in which an existing topic becomes more fine-grained by application context or process stage, interpreted as splitting. Topic proximity in the embedding space and the substantive content of representative documents is jointly considered to ensure that the results are interpreted as shifts in semantic structure rather than superficial keyword variation.

3.4.4. Topic-IPC Subclass Linkage

Finally, to connect the comparative findings to the technological classification system, we compute topic-IPC mappings by aggregating the IPC subclass distributions of documents assigned to each topic. This enables us to examine which subclasses are associated with each topic, how topic composition varies within the same subclass, and whether classification-based change, reflected in IPC share shifts, aligns with meaning-based change, reflected in topic share shifts. Furthermore, by cross-examining the topic-IPC mappings with the change-rate analysis and convergence network results from Section 3.3, we establish an empirical basis for discussing whether technological transition is characterized by the co-occurrence of changes in both classification structures and thematic structures.

4. Results

4.1. IPC-Based Technology Portfolio Analysis

4.1.1. Results of the IPC Section Distribution Analysis

Figure 2 presents the IPC section distribution, based on a single-classification rule, for 64,639 triadic patents filed between 2005 and 2024. The overall portfolio is dominated by Section C, Chemistry and Metallurgy, with 28,299 patents, accounting for 43.8 percent, followed by Section B, Performing Operations and Transporting, with 16,713 patents, accounting for 25.9 percent. Together, these two sections comprise 69.7 percent of the portfolio, suggesting that plastic recycling technologies have developed primarily on a technological base centered on chemical treatment and material conversion, as well as process-related operations including processing, separation, and transport.

Section A, Human Necessities, accounts for 6200 patents, or 9.6 percent, indicating that technology development has proceeded in parallel at a meaningful scale in downstream application and productization stages, such as packaging, containers, and consumer goods. Meanwhile, the presence of Section G, Physics, with 4491 patents, or 6.9 percent, and Section H, Electricity, with 3702 patents, or 5.7 percent, reflects an increasing integration of digital and measurement-oriented technologies with recycling processes, including quality evaluation, instrumentation, sensing, control, data processing, and equipment connectivity. Overall, while the conventional chemistry and process-centered axes represented by Sections C and B form the core of the portfolio, physics- and electricity-based enabling technologies in Sections G and H appear to be accumulating in a complementary manner to support greater precision and automation in process operation.

Sections D, Textiles, with 1901 patents, accounting for 2.9 percent, F, Mechanical Engineering, with 1829 patents, accounting for 2.8 percent, and E, Fixed Constructions, with 1504 patents, accounting for 2.3 percent, represent relatively small shares. Nevertheless, their presence indicates non-negligible technological demand in areas such as application materials for recycled feedstocks, including textiles, equipment and machinery design, and infrastructure and facility development. Overall, the technology landscape exhibits a multilayered structure comprising four main components: chemistry and composition-oriented technologies, captured by Section C; process- and operations-oriented technologies, including separation, washing, processing, and transport, captured by Section B; downstream product and application expansion, captured by Section A; and measurement, sensing, control, and connectivity-oriented enabling technologies, captured by Sections G and H. In subsequent analyses using IPC classes and subclasses as well as co-occurrence networks, the key focus is to examine how the internal axes within Sections C and B become differentiated and how technologies in Sections G and H couple with specific stages of the recycling process.

4.1.2. Results of the Annual IPC Section Distribution Analysis

Figure 3 presents a time-series of annual IPC section distributions for the period 2005 to 2024, based on adjusted counts. Throughout the entire period, Sections C, Chemistry and Metallurgy, and B, Performing Operations and Transporting, form the core of the technology portfolio, with their combined share remaining within an approximate range of 65 to 72 percent each year. This indicates that waste-plastic recycling technologies have been driven over the long run by chemical treatment and material conversion in Section C, together with process operations such as separation, washing, processing, and transport in Section B. At the same time, while these foundational axes persist, the portfolio exhibits a gradual rebalancing as cumulative shifts in section shares accrue around the period of 2015, suggesting a progressive relocation of the technological center of gravity.

The direction of change can be summarized as a relative expansion of process and operations technologies and a strengthening of digital integration. Section B increases by 5.0 percentage points and Section H, Electricity, also expands by 3.1 percentage points, whereas Section A, Human Necessities, declines by 5.6 percentage points, Section G, Physics, by 4.7 percentage points, and Section C, Chemistry and Metallurgy, by 4.5 percentage points. This pattern does not necessarily imply a weakening of chemistry-based technologies; rather, it suggests that innovation progressed more rapidly in operational stages spanning collection, sorting, treatment, and transport, resulting in a faster relative expansion of Section B. In parallel, the growth of Section H supports the interpretation that electricity and communications-based enabling technologies, including equipment control, process monitoring, and traceability and connectivity functions, are becoming increasingly embedded across the recycling value chain.

The most recent five-year window, from 2019 to 2024, shows an even clearer reallocation across sections. Section B records the largest increase, rising by 7.0 percentage points, while Section D, Textiles, expands by 2.9 percentage points, implying that application pathways for recycled feedstocks may be broadening toward textile materialization. By contrast, Sections E, Fixed Constructions, H, Electricity, F, Mechanical Engineering, and C, Chemistry and Metallurgy, decline by 3.8, 2.1, 1.7, and 2.3 percentage points, respectively. This suggests that growth in infrastructure and facility development or in machinery and chemistry-centered domains did not keep pace with the expansion in process and transport technologies, or that some functions were redistributed and absorbed into other sections during a period of adjustment. Overall, these time-series patterns indicate that, while Sections C and B maintain a dominant structural backbone over the long term, the technological system has been reconfigured since 2015 in a direction that strengthens process and operations in Section B together with connectivity and control elements in Section H. However, it is important to note that this statement requires revision considering the post-2022 trajectory observed in the graph. After reaching a peak around 2022, Section H exhibits a notable decline, suggesting that the expansion of electricity and communications-based enabling technologies may have plateaued or undergone a period of consolidation. The reasons behind this post-2022 decline in Section H warrant further discussion. Possible explanations include a maturation effect, whereby foundational control and connectivity technologies became standardized and absorbed into broader system designs, reducing the need for dedicated patent filings in this category. Alternatively, the decline may reflect a reorientation of R&D investment toward other domains, such as chemical recycling processes or logistics optimization, as the industry shifted focus following an initial wave of digitalization. Broader macroeconomic factors and shifts in patent filing strategies following the COVID-19 disruption period may also have contributed to the observed slowdown.

4.1.3. Results of the Class-Level Analysis

Figure 4 presents the IPC class distribution for the full period from 2005 to 2024, based on the top 20 classes within the total sample of 64,639 patents. Among 122 classes, the top 20 account for 76.44 percent, or 45,408 patents, indicating a portfolio structure that is highly concentrated in a limited set of core classes. Notably, G06, Computing, Calculating, and Counting, is the largest class with 9447 patents, representing 15.90 percent, followed by H04, Electric Communication Technique, with 7867 patents, representing 13.24 percent. Together, these two classes alone comprise approximately 29 percent of the portfolio, suggesting that the core locus of competition in recycling technologies has shifted strongly toward operational upgrading elements such as data processing, automation, control, and connectivity across equipment and systems, in addition to process and material innovations.

At the same time, classes including C07, Organic Chemistry, B01, Physical or Chemical Processes or Apparatus in General, C08, Organic Macromolecular Compounds, C10, Petroleum, Gas, Fuel, and Lubricants, and C12, Biochemistry and Microbiology, remain prominent, indicating that chemistry and process-based axes related to conversion, composition, separation, and reaction still form the foundational base of the technology landscape. The inclusion of G01, Measuring and Testing, H01 and H03, Basic Electric Elements and Basic Electronic Circuitry, C02, Treatment of Water, Waste Water, Sewage, or Sludge, C09, Dyes, Paints, Polishes, Resins, and Quality Modification, and B65, Conveying, Packing, and Handling, further supports the interpretation that instrumentation, quality management, wash-water and wastewater treatment, and logistics are being integrated as constituent components directly linked to process performance.

Figure 5 and Figure 6 compare changes in the composition of the top 20 IPC classes by splitting the observation window into 2005 to 2014, the earlier period, and 2015 to 2024, the later period. In the earlier period shown in Figure 5, G06 records 2680 patents, accounting for 14.77 percent, and H04 records 2248 patents, accounting for 12.35 percent, indicating that these digital and connectivity-related classes were already positioned at the top. However, the relative prominence of chemistry- and apparatus-oriented classes such as C07, with 1337 patents or 7.36 percent, and B01, with 1006 patents or 5.53 percent, is more pronounced. By contrast, in the later period shown in Figure 6, both the scale and share of the digital and connectivity axis expand, with G06 increasing to 6789 patents, or 16.48 percent, and H04 increasing to 5619 patents, or 13.64 percent, thereby strengthening their dominance. Over the same period, the share of C07 declines to 2138 patents, or 5.19 percent, while B01 increases to 2161 patents, or 5.24 percent, but its growth does not match the magnitude of the rise in the digital classes.

This pattern suggests that the locus of differentiation in technological development has shifted from novelty in chemical reaction and material pathways toward optimization, automation, connectivity, and traceability in process operations, resulting in a reconfiguration of the upper tier of the portfolio around digitally oriented classes.

In addition, the later period from 2015 to 2024 shown in Figure 6 includes stable representation of G01, Measuring and Testing, with 1038 patents or 2.52 percent; C02, Treatment of Water, Wastewater, Sewage, or Sludge, with 801 patents or 1.94 percent; C09, quality modification and compounding-related technologies, with 750 patents or 1.82 percent; and B65, logistics and handling, with 459 patents or 1.11 percent. This indicates that, as the recycling value chain expands from collection and sorting to washing, recycled-feedstock quality management, and transport and traceability, instrumentation, water treatment, and logistics are becoming embedded not merely as auxiliary functions but as core elements that shape process performance and output quality.

4.1.4. Results of the Period-by-Period Class Share Analysis

Figure 7 reports changes in the shares of the top 20 IPC classes as percentage-point differences between the later period, 2015 to 2024, and the earlier period, 2005 to 2014. Classes with large shifts separate clearly into two groups, indicating that the portfolio’s center of gravity moved in the later period toward operational upgrading, encompassing data, connectivity, and measurement, together with an outward expansion of the process boundary that incorporates washing, quality correction, and logistics.

The largest increases are observed for G06, which rises by 1.63 percentage points, and H04, which rises by 1.28 percentage points. The expansion of G06 reflects the growing weight of digitally enabled process-operation technologies, including data processing, machine-vision and AI-based sorting, process monitoring, and optimization. The increase in H04, in turn, captures the integration of IoT and communication-based connectivity that links equipment, plant operations, and logistics across the recycling system. Concurrent increases in C02, up by 0.62 percentage points, and C09, up by 0.55 percentage points, suggest that wastewater treatment and process-water reuse needs have become more salient as washing and pretreatment intensify, and that composition, additive, and modification technologies for correcting quality variability in recycled feedstocks have emerged as more important constraints in the later period. Increases in H01, up by 0.56 percentage points, and G01, up by 0.15 percentage points, indicate a strengthening of measurement and control enabling technologies, such as electromechanical and power-control components as well as sensing, inspection, and analytical functions, to stabilize operations and secure product quality. B65 also rises by 0.10 percentage points, consistent with a gradual expansion of collection, transport, and storage infrastructure within the portfolio.

By contrast, the decline group is led by a pronounced decrease in C07, down by 2.16 percentage points. This suggests that, while the organic-synthesis–oriented conversion axis remains active, its relative prominence within the upper portfolio has fallen because technologies related to process operations, quality, and system integration grew more rapidly in the later period. Decreases in H03, down by 0.97 percentage points, and C10, down by 0.46 percentage points, are consistent with a reallocation from traditional circuit-centric approaches toward communications, data processing, and system connectivity captured by H04 and G06, and with fuelization and conversion-oriented pathways not keeping pace with the growth of operations, quality management, and materialization-oriented axes. Declines in B01, down by 0.28 percentage points, C12, down by 0.26 percentage points, and C08, down by 0.12 percentage points, likewise are better interpreted as relative reweighting rather than absolute contraction, as digitally enabled sorting and control and the expansion of peripheral process functions, including washing and water treatment, quality correction, and logistics, accelerated in the later period.

These share shifts substantiate, at the class level, the section-level pattern observed in the previous subsection, namely the strengthening of Section B and the expansion of Section H. In other words, while the chemistry and process core, represented by classes such as C07, C08, and B01, remains foundational, the later period exhibits a portfolio reconfiguration in which operations-centered structures combining data and connectivity in G06 and H04, measurement and control in G01 and H01, auxiliary process functions in C02 and C09, and collection and logistics in B65 increasingly shape the upper tier of the technology landscape.

4.1.5. Results of the Subclass-Level Analysis

Table A1. Subclass-level analysis.

Level	Code	Name	Section	Parent_Class
subclass	G06F	Electric digital data processing	G	G06
subclass	G06N	Computer systems based on specific computational models	G	G06
subclass	G06V	Image or video recognition or understanding	G	G06
subclass	G06T	Image data processing or generation, in general	G	G06
subclass	G06K	Recognition of data; presentation of data; record carriers; handling record carriers	G	G06
subclass	G06Q	Data processing systems or methods, specially adapted for administrative, commercial, financial, managerial, supervisory or forecasting purposes	G	G06
subclass	G06G	Analog computers	G	G06
subclass	G06M	Computing arrangements for specific functions	G	G06
subclass	H04W	Wireless communication networks	H	H04
subclass	H04L	Transmission of digital information, e.g., telegraphic communication	H	H04
subclass	H04J	Multiplex communication	H	H04
subclass	H04B	Transmission	H	H04
subclass	H04N	Pictorial communication, e.g., television	H	H04
subclass	H04K	Secret communication; jamming of communication	H	H04
subclass	H04M	Telephonic communication	H	H04
subclass	H04Q	Selecting	H	H04
subclass	C02F	Treatment of water, waste water, sewage, or sludge	C	C02
subclass	H01M	Processes or means, e.g., batteries, for the direct conversion of chemical energy into electrical energy	H	H01
subclass	H01J	Electric discharge tubes or discharge lamps	H	H01
subclass	H01R	Electrically conductive connections; structural associations; coupling devices; current collectors	H	H01
subclass	H01G	Capacitors; capacitors and condensers	H	H01
subclass	H01B	Cables; conductors; insulators; selection of materials	H	H01
subclass	H01L	Semiconductor devices; electric solidstate devices not otherwise provided for	H	H01
subclass	H01T	Spark gaps; overvoltage arresters; voltage limiters; discharge tubes	H	H01
subclass	H01S	Devices using stimulated emission	H	H01
subclass	C09K	Materials for applications not otherwise provided for; compositions thereof	C	C09
subclass	C09J	Adhesives; non-mechanical aspects of adhesive processes, in general	C	C09
subclass	C09C	Treatment of inorganic materials, other than fibrous fillers, to enhance their pigmenting or filling properties	C	C09
subclass	C09G	Polishing compositions other than French polish; ski waxes	C	C09
subclass	C09D	Coating compositions, e.g., paints, varnishes, lacquers; printing inks; etc.	C	C09
subclass	C09B	Organic dyes or closely related compounds for producing dyes; mordants; lakes	C	C09
subclass	C09F	Obtaining, purification, or chemical modification of natural resins	C	C09
subclass	C09H	Preparation of glue or gelatine	C	C09
subclass	A61B	Diagnosis; surgery; identification	A	A61
subclass	A61K	Preparations for medical, dental, or toilet purposes	A	A61
subclass	A61L	Methods or apparatus for sterilizing materials or objects; disinfection; etc.	A	A61
subclass	A61J	Medical/pharmaceutical containers; administering forms; etc.	A	A61
subclass	A61G	Transport or accommodation for patients; assisting disabled persons	A	A61
subclass	A61F	Prostheses; orthopedic, nursing or contraceptive devices; bandages; etc.	A	A61
subclass	A61M	Devices for introducing media into, or onto, the body; etc.	A	A61
subclass	A61C	Dentistry; apparatus or methods for oral or dental hygiene	A	A61
subclass	G01N	Investigating or analyzing materials by determining chemical or physical properties	G	G01
subclass	G01K	Measuring temperature; measuring quantities of heat	G	G01
subclass	G01S	Radio direction-finding; radio navigation; locating by radio waves	G	G01
subclass	G01F	Measuring volume, volume flow, mass flow, or liquid level	G	G01
subclass	G01B	Measuring length, thickness, angles, areas; measuring surface irregularities	G	G01
subclass	G01V	Geophysics; detecting masses or objects; etc.	G	G01
subclass	G01C	Surveying; navigation; gyroscopic instruments; photogrammetry/videogrammetry	G	G01
subclass	G01M	Testing of structures or apparatus not otherwise provided for	G	G01
subclass	E21B	Earth or rock drilling; obtaining oil, gas, water, etc. from wells	E	E21
subclass	E21C	Mining or quarrying	E	E21
subclass	E21D	Shafts; tunnels; galleries; underground chambers	E	E21
subclass	E21F	Safety devices; transport; rescue; ventilation in mines or tunnels	E	E21
subclass	B65D	Containers for storage or transport of articles or materials	B	B65
subclass	B65F	Gathering or removal of refuse; refuse transport; sorting; disposal	B	B65
subclass	B65G	Transporting; conveying; storing	B	B65
subclass	B65B	Packaging machines/apparatus; unpacking	B	B65
subclass	B65H	Handling thin or filamentary material	B	B65
subclass	B65C	Labeling or tagging machines, apparatus, or processes	B	B65
subclass	F01D	Non-positive-displacement machines or engines, e.g., steam turbines	F	F01
subclass	F01N	Gas-flow silencers or exhaust apparatus	F	F01
subclass	F01P	Cooling of machines or engines in general	F	F01
subclass	F01L	Cyclically operating valves for machines or engines	F	F01
subclass	F01B	Reciprocating machines or engines; steam engines	F	F01
subclass	F01M	Lubricating of machines or engines in general	F	F01
subclass	F01C	Rotary-piston machines or engines	F	F01
subclass	F01K	Steam engine plants; engine plants not otherwise provided for	F	F01
subclass	H02M	Apparatus for conversion between ac and dc or between dc and dc; power supplies	H	H02
subclass	H02K	Dynamo-electric machines	H	H02
subclass	H02S	Generation of electric power by conversion of light, e.g., photovoltaic	H	H02
subclass	H02J	Circuit arrangements or systems for supplying or distributing electric power; energy storage	H	H02
subclass	H02H	Emergency protective circuit arrangements	H	H02
subclass	H02N	Electric machines not otherwise provided for	H	H02
subclass	H02P	Control or regulation of electric motors, generators, or converters	H	H02
subclass	H02G	Installation of electric cables or lines	H	H02
subclass	C08J	Working-up; general processes of compounding; after-treatment	C	C08
subclass	C08L	Compositions of macromolecular compounds	C	C08
subclass	C08H	Derivatives of natural macromolecular compounds	C	C08
subclass	C08F	Macromolecular compounds obtained by reactions only involving C=C unsaturated bonds	C	C08
subclass	C08B	Polysaccharides; derivatives thereof	C	C08
subclass	C08G	Macromolecular compounds obtained otherwise than by reactions only involving C=C unsaturated bonds	C	C08
subclass	C08C	Treatment or chemical modification of rubber	C	C08
subclass	C08K	Use of inorganic or non-macromolecular organic substances as compounding ingredients	C	C08
subclass	F02M	Supplying combustion engines with combustible mixtures or constituents thereof	F	F02
subclass	F02D	Controlling combustion engines	F	F02
subclass	F02N	Starting of combustion engines	F	F02
subclass	F02F	Cylinders, pistons, casings for combustion engines	F	F02
subclass	F02C	Gas-turbine plants; air-intakes for jet-propulsion plants; etc.	F	F02
subclass	F02B	Internal-combustion piston engines; combustion engines in general	F	F02
subclass	F02G	Hot-gas or combustion-product positive-displacement engine plants	F	F02
subclass	F02K	Jet-propulsion plants	F	F02
subclass	A23N	Machines or apparatus for treating harvested fruit/vegetables in bulk	A	A23
subclass	A23F	Coffee; tea; substitutes; manufacture or infusion	A	A23
subclass	A23L	Foods or non-alcoholic beverages not covered by A23B-A23J; preparation or treatment	A	A23
subclass	A23J	Protein compositions for foodstuffs; working-up proteins; phosphatide compositions	A	A23
subclass	A23C	Dairy products; substitutes; making thereof	A	A23
subclass	A23G	Cocoa; chocolate; confectionery; ice-cream	A	A23
subclass	A23K	Fodder	A	A23
subclass	A23B	Preserving meat/fish/fruit/vegetables; chemical ripening; preserved products	A	A23
subclass	C12Q	Measuring or testing processes involving enzymes, nucleic acids, or microorganisms	C	C12
subclass	C12N	Microorganisms or enzymes; compositions; genetic engineering	C	C12
subclass	C12F	Beer; preparation thereof	C	C12
subclass	C12G	Wine; other alcoholic beverages; preparation thereof	C	C12
subclass	C12H	Pasteurization/sterilization/preservation/purification of alcoholic beverages	C	C12
subclass	C12P	Fermentation or enzyme-using processes to synthesize a chemical compound or composition	C	C12
subclass	C12C	Brewing of beer	C	C12
subclass	C12M	Apparatus for enzymology or microbiology	C	C12
subclass	B01L	Chemical or physical laboratory apparatus for general use	B	B01
subclass	B01J	Chemical or physical processes, e.g., catalysis; colloid chemistry; apparatus	B	B01
subclass	B01D	Separation	B	B01
subclass	B01F	Mixing, e.g., dissolving, emulsifying, dispersing	B	B01
subclass	C01F	Compounds of Be, Mg, Al, Ca, Sr, Ba, Ra, Th, or rare-earth metals	C	C01
subclass	C01B	Non-metallic elements; compounds thereof	C	C01
subclass	C01G	Compounds containing metals not covered by C01D or C01F	C	C01
subclass	C01D	Compounds of alkali metals	C	C01
subclass	C01C	Ammonia; cyanogen; compounds thereof	C	C01
subclass	C01J	Inorganic materials or general use; compositions thereof	C	C01
subclass	C10B	Destructive distillation of carbonaceous materials for production of gas, coke, tar, etc.	C	C10
subclass	C10L	Fuels not otherwise provided for; natural gas; additives to fuels; etc.	C	C10
subclass	C10J	Production of producer gas, water gas, synthesis gas from solid carbonaceous materials	C	C10
subclass	C10G	Cracking hydrocarbon oils; production of liquid hydrocarbon mixtures; etc.	C	C10
subclass	C10M	Lubricating compositions	C	C10
subclass	C10C	Working-up of tar, pitch, asphalt, bitumen	C	C10
subclass	C10K	Purification or modification of gaseous fuels	C	C10
subclass	C10H	Production of acetylene or similar unsaturated hydrocarbons	C	C10
subclass	H03L	Automatic control, starting, synchronization, or stabilization of generators of oscillations or pulses	H	H03
subclass	H03M	Coding, decoding or code conversion	H	H03
subclass	H03D	Demodulation or transference of modulation	H	H03
subclass	H03K	Pulse technique	H	H03
subclass	H03H	Impedance networks; resonators	H	H03
subclass	H03B	Generation of oscillations	H	H03
subclass	H03C	Modulation	H	H03
subclass	H03F	Amplifiers	H	H03
subclass	C07G	Compounds of unknown constitution	C	C07
subclass	C07C	Acyclic or carbocyclic compounds	C	C07
subclass	C07D	Heterocyclic compounds	C	C07
subclass	C07F	Acyclic/carbocyclic/heterocyclic compounds containing elements other than typical set	C	C07
subclass	C07H	Sugars; derivatives; nucleosides/nucleotides/nucleic acids	C	C07
subclass	C07J	Steroids	C	C07
subclass	C07K	Peptides	C	C07
subclass	C07B	General methods of organic chemistry; apparatus	C	C07

in the Appendix A provides a more fine-grained view of which specific functions underpin the class-level structural changes identified in the previous subsection. Subclasses that increase toward the later period from 2015 to 2024 can be summarized along two main axes. The first axis concerns data processing and recognition-driven upgrading of process operations within the G06 family. Electrical digital data processing in G06F increases by 1.05 percentage points, and computing based on specific computational models in G06N rises by 0.57 percentage points. Image and video recognition in G06V and image data processing in G06T also increase by 0.23 and 0.21 percentage points, respectively. This pattern suggests that sorting and quality management, repeatedly identified as key bottlenecks in recycling processes, are being strengthened through automation and intelligentization based on sensor and image data. By contrast, some subclasses related to recording media and data handling in G06K decline by 0.25 percentage points, and data processing for administration, supervision, or forecasting purposes in G06Q fall by 0.11 percentage points. This contrast indicates that digital transformation in this domain has progressed less through administrative or commerce-oriented data processing and more through recognition, processing, and computation functions that couple directly to on-site operations.

The second axis relates to integrated operations enabled by on-site connectivity and wireless networking within the H04 family. Wireless communication networks in H04W increase markedly by 1.78 percentage points, and digital information transmission in H04L rises by 1.01 percentage points, whereas traditional communication subdomains such as multiplex communication in H04J, general transmission in H04B, and telephonic communication in H04M decline by 0.54, 0.32, and 0.13 percentage points, respectively. This contrast implies a reallocation in which demand grows for IoT and network-oriented communication functions that connect equipment, production lines, logistics, and traceability across the recycling value chain, while the relative weight of general-purpose communication technologies decreases. In other words, the shift reflects less an expansion in the variety of communication technologies than a selective strengthening of network functions required for real-time transmission and the interoperability of process data.

An outward expansion of the process boundary is also evident at the subclass level. The rise in water treatment in C02F, up by 0.62 percentage points, reflects increasing environmental and operational requirements, including process-water reuse and pollutant-load reduction, alongside intensified washing, and pretreatment. Increases in composition- and materials-related subclasses linked to quality correction, such as C09K, up by 0.38 percentage points, and adhesive and processing-related functions in C09J, up by 0.13 percentage points, indicate a strengthening emphasis on post-treatment and formulation optimization to correct property variability in recycled feedstocks and to meet application-specific requirements.

By contrast, declining subclasses point to a relative weakening of traditional chemical conversion and separation unit operations. Within the C07 domain, the subclass associated with acyclic or carbocyclic compounds, C07C, decreases substantially by 1.65 percentage points, suggesting that, although synthesis- and reaction-centered conversion activities remain, they did not keep pace with the growth of operations, quality, and connectivity-oriented axes in the later period. Separation processes in B01D also decline by 0.38 percentage points, which is more plausibly interpreted as a relative reweighting under the rapid expansion of digitally enabled sorting and quality management in areas such as G06V and G06T as well as analytical functions in G01N, and system integration functions in H04W and H04L, rather than as an absolute contraction of separation technology development. Within the fuel and gas domain in C10, subclasses such as fuels in general in C10L and gas production including synthesis gas in C10J decrease by 0.23 and 0.14 percentage points, respectively, indicating that fuelization and conversion-oriented pathways have moved to a relatively lower priority under the recent growth phase dominated by operations and quality considerations. Within the H03 domain, declines in coding and decoding in H03M and modulation and demodulation in H03D, down by 0.36 and 0.35 percentage points, align with a broader reallocation from traditional circuit- and signal-processing functions toward network-based connectivity captured by H04W and H04L.

Overall, the subclass-level shifts reported in

Table 2. Topic Transition Map and Thematic Change.

Early Period ID	Early Period Topic Label	Status	Later-Period Correspondence or Destination
E1	Thermal energy recovery and flow	Split	L2, L1
E2	Optical networking and circuits	Shrink or absorbed	L1, partial
E3	Fluid and surface and reuse	Reordered as applications	L3, L4
E4	Polymer coating and polymerization	Reordered as applications	L4, L6
E5	Nucleic acid extraction, bio	Split or converged	L5, L7
E6	Optical imaging and display	Refined	L5
E7	Semiconductor wafer processing	Split toward precision·reliability	L10, L11
E8	Inorganic quartz and minerals	Disappeared	None
E9	Circuit design and tessellation	Disappeared	None
E10	Nuclear fuel and fission	Disappeared	None

decompose the class-level expansion of G06 and H04 into concrete functional changes, namely AI- and computer-vision-based sorting and quality management and wireless-network and digital-transmission-based connectivity and traceability. At the same time, increases in water treatment in C02F and quality correction in C09K and C09J support the interpretation that waste-plastic recycling technologies are evolving beyond conversion technologies alone toward configurations that simultaneously satisfy operational performance, environmental constraints, and quality standardization requirements.

4.2. Topic Evolution and IPC Mapping Analysis

4.2.1. Bert Topic Modeling

Figure 8 visualizes, in a two-dimensional space, the embedding-based distribution of 10 topics derived using BERTopic for the 2005 to 2014 period. As points belonging to the same topic form dense clusters and greater inter-topic distances indicate lower lexical and contextual similarity, the map suggests that documents in this period are organized into several relatively separated thematic groups. The largest clusters are dominated by process- and material-oriented themes such as thermal energy recovery, fluid surface processing, and polymer coating. By contrast, optical networking and optical imaging form distinct clusters, indicating that instrumentation and communication elements related to process operation existed as independent themes alongside core process topics.

In addition, topics such as semiconductor wafer processing, inorganic quartz, circuit design, nuclear fuel systems, and nucleic acid extraction appear as separate clusters with weak direct linkage to the recycling domain. This pattern can be interpreted as a signal that the dataset is not perfectly confined to plastic recycling alone, and that a broader set of technical documents related to processes, materials, and equipment may be co-captured. Accordingly, the topic map for this period serves to visually confirm a structure in which core process- and material-centered themes coexist with adjacent-industry or general-purpose process and materials themes.

Figure 9 presents word scores for the most representative keywords within each topic, providing a basis for topic interpretation. Thermal energy recovery, Topic 1, is characterized by energy- and fluid-mediated terms such as gas, water, stream, heat, and recovery, forming a context related to heat recovery, utilization of thermal sources, and the recovery or reuse of process fluids. Optical networking, Topic 2, is characterized by clock, network, storage, recovery, and optical, indicating a focus on communication and networking functions such as data transmission, synchronization, and storage. Fluid surface processing, Topic 3, features terms such as fluid, ink, surface, and reusable, capturing contexts that link coating or surface-treatment processes with reusability or surface functionalization in recycling-related operations. Polymer coating, Topic 4, is represented by coating, compound, polymer, material, and polymerization, consistent with a materials-oriented theme centered on polymer formulation, polymerization, and film formation.

For topics that are less specific to the recycling domain, the keyword sets align with typical technical vocabularies of those fields and thereby support the validity of the assigned labels. Nucleic acid extraction, Topic 5, includes terms such as nucleic, silicon, treatment, and extract, indicating a strong bio-analytical and treatment-related context. Optical imaging, Topic 6, includes image, light, display, and apparatus, reflecting an optical-device and image-generation context. Semiconductor wafer processing, Topic 7, is characterized by semiconductor, wafer, silicon, and processing, while inorganic quartz, Topic 8, is dominated by titanium, quartz, sodium, and sio2, suggesting an independent theme focused on inorganic material composition and feedstocks. Circuit design, Topic 9 includes design and circuit, as well as terms such as tessellation and constraint that reflect optimization and structuring in design contexts. Nuclear fuel systems, Topic 10 is dominated by fuel, fission, uranium, and fissile, which are typical keywords of nuclear fuel and fission technologies.

Overall, the topic structure for 2005 to 2014 is centered on process- and material-oriented themes related to heat, fluids, and coatings, while also exhibiting the coexistence of separate clusters drawn from adjacent technological domains such as optics, communications, semiconductors, inorganic materials, biotechnology, and nuclear fuel. In comparison with the 2015 to 2024 results reported in the subsequent subsection, the analysis focuses on whether topics directly linked to operational upgrading, including sorting, measurement, control, and connectivity, become more prominent; how the relative weight and cluster cohesion of domain-nonspecific topics change; and how topics mapped to subclasses associated with G06, H04, and G01N are reconstituted in terms of representative keyword structures.

Figure 10 visualizes the embedding-based distribution of 11 topics derived using BERTopic for the 2015 to 2024 period in a two-dimensional space. In the visualization, topic groups directly related to recycling appear in a relatively cohesive form. Reverse logistics themes, including take-back, washing, labeling, and refill for reusable packaging, Topic 1; remanufacturing, Topic 2; chemical recycling for packaging, Topic 3; thermal and fluid-stream-based energy and process operation, Topic 4; system recovery and safety, Topic 5; and automation and data sources, Topic 10, are located near the central area. This suggests that, in the later period, the thematic structure of recycling technologies was reorganized from a focus on single conversion processes toward a value-chain perspective linking collection, operations, quality, safety, and data. At the same time, topics such as radiation patterning, ferroelectric fatigue, sequence imaging, robotic surgery scaffolds, and hybrid vehicle technologies remain separated as themes with weak direct linkage to the recycling domain, indicating that broad process, materials, and equipment-related documents can still be co-captured in this period.

Figure 11 reports word scores for representative keywords of each topic, strengthening the basis for topic labeling. Reverse logistics and reusable packaging, Topic 1, is characterized by sorting, label, return, refill, and reusable, forming an operational context centered on take-back, classification, return, and refill. Chemical recycling for packaging, Topic 3, is characterized by recycled, supply, glycolysis, packaging, and methanolysis, clearly reflecting chemical recycling pathways centered on depolymerization and conversion routes such as glycolysis and methanolysis. Thermal and fluid streams, Topic 4, are represented by gas, stream, heat, material, and water, indicating a focus on process fluids, heat sources, and operating conditions. Remanufacturing, Topic 2, includes remanufacturing, repairability, refurbishment, and recovery, representing life-extension circular strategies such as repairability and refurbishment. Data sources, Topic 10, features source, properties, coding, statistical, and metadata, indicating an analytical-infrastructure theme related to the collection, coding, and statistical processing of process and quality data.

The topic configuration in the later period is meaningful in that themes directly connected to the recycling value chain become more clearly articulated than in the earlier period. The simultaneous presence of reverse logistics, including take-back, labeling, and refill; chemical recycling, including depolymerization and conversion; remanufacturing as a life-extension strategy; thermal and fluid-based operation; system recovery and safety; and data-oriented analytical infrastructure supports the interpretation that the focus of technological development has expanded beyond upgrading unit process technologies toward integrating operational systems, including traceability, data, and automation, and managing quality and safety constraints.

4.2.2. Topic Change Based on the Transition Map

The topic transition visualization map in Table 2 aligns topics from the early period, 2005 to 2014, with those from the later period, 2015 to 2024, within a shared embedding space. The mapped movements indicate that the thematic structure was not simply replaced, but was reorganized through processes of refinement, recombination, and disappearance. In the early period, process- and materials-oriented themes such as thermal energy recovery and fluid flow, fluid surface processing, and polymer coating and polymerization form the focus. In contrast, the later period shows strengthened themes that are more tightly coupled to circular value chains and operational systems, including reverse logistics for reusable packaging, remanufacturing, chemical recycling of packaging, system recovery, and data sources.

At a more granular level, E1 connects to L2, thermal and fluid streams, and L1, system recovery and stability, suggesting that attention to energy recovery expanded toward operational conditions and system reliability. E2, optical networking and circuits, is only partially absorbed into L1, indicating a reconfiguration in which emphasis shifted away from communications and circuit technologies per se and toward connectivity functions that couple directly to process operation. E3 is linked to L3, remanufacturing, and L4, operations and reuse, implying that enabling technologies related to surfaces, cleaning, and contamination management were recontextualized toward life-extension strategies and take-back and reuse systems. E4 is reassigned to L4 and L6, suggesting a shift in emphasis from coating and polymerization toward conversion and regeneration pathways, including depolymerization. By contrast, several topics with stronger adjacent-industry characteristics, such as E8 through E10, show weak correspondence or disappearance in the later period, indicating a relative strengthening of recycling-centered topics after 2015. These movements are also consistent with the IPC findings of expanded process and operations technologies in Section B and increased data and connectivity functions captured by G06 and H04, supporting the interpretation that the later-period technology structure was reconfigured toward a systems-level perspective.

4.2.3. Comparative Interpretation

Figure 12 schematically aligns the BERTopic results for 2005 to 2014 and 2015 to 2024 to illustrate pathways of topic evolution, extinction, and emerging themes. Solid links indicate semantic continuation or transformation from an early-period topic to a later-period topic, dashed links indicate the relative extinction of early-period topics, and green arrows with the “new field” label denote themes that emerge as independent topics in the later period. The diagram indicates that thematic change proceeded not as simple replacement, but through the simultaneous progression of continuity in selected themes, function-centered redefinition, and the emergence of new thematic clusters driven by value-chain expansion.

Core themes in the early period consisted largely of process, materials, and device-centered topics, including thermal energy recovery, fluid and surface processing, coating and polymerization, and optical imaging and networking. Within these, the transition from Thermal energy recovery to Thermal fluid streams suggests that a focus on energy recovery expanded toward an operations and management perspective that incorporates thermal and fluid streams within the process. The transition from Optical imaging to Sequence imaging indicates that measurement and imaging functions shifted toward more advanced and precision-oriented analytical domains. The transition from Optical networking to Data sources implies that connectivity technologies became less of an end in themselves, while data-oriented operational infrastructure such as collection, coding, and the metadata management of process and quality data rose as an independent technological axis. Material and surface-oriented topics such as Polymer coating and Fluid surface processing link to later-period themes including Chemical recycling for packaging and System recovery, which can be interpreted as a reallocation from materials functionality toward system-level performance goals such as the conversion and regeneration of post-consumer packaging through depolymerization and related pathways, as well as process reliability and recovery.

In contrast, topics such as Nucleic acid extraction, Inorganic quartz, and Nuclear fuel systems show weak direct inheritance or are depicted as extinct, while operational and strategic themes directly tied to circular-economy implementation, such as Reusable packaging logistics and Remanufacturing, appear as “new field” topics. This indicates that the later-period thematic structure expanded beyond unit-process technologies toward an integrated value-chain perspective linking take-back, reuse, remanufacturing, and conversion.

4.2.4. Link Topics to IPC Subclasses

Table 3. Link topics to IPC subclasses.

No	Topic (A→B)	Increase (+)	Decrease (−)	Neutral/Other
1	Thermal energy recovery → Thermal fluid streams		C10L, C10J	F01
2	Optical imaging → Sequence imaging	G06V, G06F, G01N		A61
3	Semiconductor wafer processing → High-resolution radiation patterning	G06F	C07C	H01, C08
4	Semiconductor wafer processing → Ferroelectric fatigue	G06F	H03M, H03D, H03K	H01
5	Polymer coating → Chemical recycling for packaging	C08J, B65D, B65F, B65G C09K, C09J, C02F	C07C
6	Circuit design → System recovery	G06F, H04W, H04L	H03M, H03D, H03K
7	Fluid surface processing → High-resolution radiation patterning	C09K, C09J, G06F		H01, C08
8	Optical networking → Data sources	H04W, H04L G06F, G06N
9	Nucleic acid extraction → —	G01N	C12P	A61
10	Inorganic quartz → —			C01, H01
11	Nuclear fuel systems → —			G21
12	Reusable packaging logistics	B65D, B65F, B65G, H04W H04L, G06F, G06V
13	Remanufacturing	B65D, B65F, B65G, H04W H04L, G06F, G01N, C08J
14	Robotic tissue scaffolds	G06N, G01N	C12P	A61
15	Hybrid vehicle technologies	H04W, H04L, G06F		H02, F01

summarizes changes in the shares of IPC subclasses associated with each transition pathway from early-period topics to later-period topics as depicted in Figure 12. A plus sign indicates technological elements whose relative weight increases in the later period, a minus sign indicates elements that decrease, and neutral or other categories indicate elements that are linked to the transition without a clear directional change. The table shows that topic-level semantic change is structurally coupled with operational upgrading based on data, connectivity, and measurement in subclasses within G06, H04, and G01; with logistics and collection functions within B65; and with process-peripheral functions such as water treatment in C02F and quality correction in C09-related subclasses.

Across transitions directly connected to the recycling value chain, strengthening digital connectivity and operational infrastructure is repeatedly observed. The transition from Optical imaging to Sequence imaging is associated with increased shares of image recognition in G06V, digital data processing in G06F, and analytical and inspection functions oriented toward material and contamination identification in G01N, indicating that measurement, discrimination, and data processing became core functions in the later-period topic structure. Similarly, the transition from Optical networking to Data sources is accompanied by increases in wireless networking in H04W and digital information transmission in H04L, together with increases in G06F and G06N, supporting the interpretation that connectivity technologies evolved into operational systems centered on data collection and processing rather than constituting an end goal. The transition from Circuit design to System recovery is also accompanied by increases in G06F and in H04W and H04L, while traditional signal-processing subclasses such as H03M, H03D, and H03K decline, indicating that network-based interoperability and system recovery and reliability functions became more important than circuit- or signal-centric functions in the later period.

Transitions related to chemical recycling and the expansion of the process boundary show concurrent strengthening of post-treatment, quality correction, water treatment, and logistics. The transition from Polymer coating to Chemical recycling for packaging is associated with simultaneous increases in polymer post-treatment and compounding in C08J, logistics and take-back and transport in B65D, B65F, and B65G, formulation and additive-based quality correction in C09K and related C09 subclasses, and water treatment in C02F. This indicates that later-period chemical recycling is not sustained by reaction and conversion technologies alone, but expands in combination with an operational system spanning take-back, sorting, washing and water treatment, property correction, and transport and handling. In contrast, the same transition shows decreases in C07C, a key organic-chemistry subclass, suggesting that synthesis- and reaction-centered traditional organic chemistry becomes relatively less prominent.

The same pattern is reinforced in newly emerging topics. Reusable packaging logistics and Remanufacturing display common increases in B65-related collection and logistics functions, H04-related connectivity, and G06-related data processing and recognition, indicating that later-period emerging themes are formed because of operations and logistics integrated with digital capabilities. Remanufacturing is additionally associated with increased material and quality identification functions in G01N and selected materials processing in C08J, implying that life-extension circular strategies are implemented through the integration of quality assessment, process treatment, and logistics systems. By contrast, adjacent-domain topics such as Nucleic acid extraction and Robotic tissue scaffolds are associated with decreases in C12P and neutral patterns in medical subclasses such as A61, suggesting that these shifts are better interpreted as structural adjustments of adjacent-industry topics within the dataset rather than as changes directly coupled to the recycling domain.

In summary, the later-period reconfiguration of the technology structure can be explained by the concurrent strengthening of data processing, recognition, and inspection functions, including G06F, G06V, G06N, and G01N; wireless and digital-transmission-based connectivity functions, including H04W and H04L; collection, transport, and storage functions, including B65D, B65F, and B65G; washing and water-treatment functions, including C02F; and formulation and quality-correction functions, including C09K and C09. In other words, the subclasses that repeatedly appear as increasing elements along topic-transition pathways reaffirm, at a micro-level, the operations- and systems-centered expansion trend identified in the IPC portfolio analysis, and they provide empirical support for the interpretation that, since 2015, plastic recycling technologies have shifted from a focus on unit process technologies toward value-chain operations integrated with digital capabilities.

5. Discussion

This study set out to characterize structural change in plastic recycling innovation by comparing patent portfolios, convergence networks, and thematic systems across the pre- and post-2015 periods using 64,639 triadic patents. The three research objectives were to (1) identify IPC portfolio shifts, (2) assess changes in convergence structure using co-occurrence networks, and (3) extract topic transitions via BERTopic modeling. The results consistently and coherently address all three objectives and converge on a unified interpretation: since 2015, the center of gravity of patent activity in plastic recycling has shifted away from chemistry-centered conversion processes toward a functionally integrated operational system spanning collection, sorting, quality correction, traceability, and process optimization. The discussion below interprets these findings in relation to the three research questions and situates them within the existing literature on circular economy technology transitions, digital transformation in waste management, and patent-based innovation analysis.

Addressing RQ1, the IPC portfolio analysis reveals a clear and quantifiable structural shift. At the section level, Sections C (Chemistry) and B (Performing Operations) together accounted for approximately 70% of the portfolio throughout 2005–2024, confirming that chemistry-based conversion and process operations form the enduring technological backbone of the field. However, the inter-period share analysis shows that Section B expanded by 5.0 percentage points and Section H (Electricity) by 3.1 percentage points, while Section C declined by 4.5 percentage points. This pattern directly addresses RQ1 by showing that the composition of the portfolio shifted in favor of operational and connectivity functions, not merely growing in absolute size. At the class level, G06 (Computing) recorded the largest share increase (+1.63 pp) and was the single largest class at 15.90% of the full-period portfolio, while H04 (Communications) ranked second (+1.28 pp, 13.24%). Together these two classes account for approximately 29% of all patent activity, a concentration that indicates the locus of R&D differentiation has shifted decisively toward data processing, AI-enabled sorting, wireless connectivity, and traceability. Increases in C02F (+0.62 pp, water treatment) and C09K (+0.38 pp, quality correction) further indicate that the process boundary of recycling innovation has expanded beyond conversion reactions to encompass washing, contamination management, and property standardization. Declines in C07 (−2.16 pp) are best interpreted as relative reweighting under faster growth in digital domains rather than absolute contraction of organic chemistry activity.

Addressing RQ2, the IPC co-occurrence network analysis reveals a qualitative reconfiguration of convergence structure, not merely quantitative growth. In the early period (2005–2014), the dominant co-occurrence combinations centered on polymer formulation, synthesis, and compounded reflecting a technology system oriented around material performance and chemical conversion pathways. In the later period (2015–2024), this formulation-centered core was supplemented and, in network centrality terms, partially displaced by a new cluster of linkages coupling post-treatment and standardization functions with processing, logistics (B65-class subclasses), and water treatment (C02F). These findings directly address RQ2 by documenting that convergence has shifted from material-chemistry combinations toward cross-domain linkages that integrate process operations with quality management, environmental constraints, and logistical infrastructure, precisely the combination profile required for a functional recycling value chain. This structural finding aligns with circular economy literature’s argument that recycling bottlenecks are systemic rather than limited to individual unit processes [7,8], and extends it by providing patent-based quantitative evidence of which specific inter-technology combinations have strengthened since 2015.

Addressing RQ3, the BERTopic analysis reveals that the semantic structure of the patent corpus underwent a substantive reorganization rather than incremental expansion. In the early period, the dominant topics were generic process- and materials-oriented (thermal energy recovery, polymer coating, fluid surface processing), with only partial linkage to the recycling domain. In the later period, the most prominent themes—chemical recycling for packaging, reverse logistics for reusable packaging, remanufacturing, and data sources—are directly coupled to circular economy value-chain stages and to the operational infrastructure supporting them. The topic-transition mapping further documents that this reorganization followed systematic pathways: thermal energy recovery expanded toward fluid-stream operations; optical networking was absorbed into data-source infrastructure; and polymer coating and surface-processing themes were redirected toward chemical recycling and system recovery topics. These transitions are not arbitrary: each maps onto IPC subclasses that expanded in share terms in the portfolio analysis (G06F, G06N, H04W, C08J, C02F), confirming that semantic and structural change are co-occurring dimensions of the same underlying transition. The finding that several adjacent-industry topics (e.g., nuclear fuel systems, inorganic quartz) disappeared or weakened in relative prominence in the later period further supports the interpretation that the post-2015 portfolio became more tightly focused on recycling-relevant innovation domains.

From a corporate R&D and IP strategy perspective, the IPC class-level results provide several directional signals grounded directly in the patent evidence. The largest portfolio share increases between the two periods are recorded for G06 (+1.63 percentage points) and H04 (+1.28 percentage points), driven at the subclass level by G06F (+1.05 pp), G06N (+0.57 pp), and G06V (+0.23 pp) within the data processing and AI domain, and by H04W (+1.78 pp) and H04L (+1.01 pp) within the connectivity domain. These figures indicate that the areas of greatest patent activity concentration in plastic recycling have shifted toward AI and vision-based sorting, computational process optimization, wireless networking, and digital data transmission. Firms can use these portfolio shift patterns as a reference signal when evaluating whether their own R&D and IP coverage is aligned with where innovation activity has demonstrably concentrated. The concurrent increases in C02F (+0.62 pp, water treatment), C09K (+0.38 pp, quality correction), and C09J (+0.13 pp, formulation) further indicate that feedstock quality management and process-water handling have become structurally more prominent across the portfolio, representing functional areas where assessing internal capability gaps may be relevant for firms designing new processes or entering new material streams.

From a policy perspective, the co-occurrence network finding that cross-domain linkages involving standardization, logistics, and water treatment strengthened in the later period implies that the bottlenecks are systemic not confined to individual technologies but spanning actor boundaries. Policy support directed at sorting infrastructure, traceability and data standards, quality measurement and certification frameworks, and process-water criteria would therefore be targeted at the functional domains where patent-based evidence of concentrated R&D effort is strongest.

The patent portfolio shifts identified in this study are broadly consistent with observable market-level developments, though direct causal inference from patent data alone is not possible. The post-2015 expansion of G06 and H04 subclasses corresponds to a period in which technology-driven small and medium-sized enterprises (SMEs) and start-ups specializing in AI-enabled recycling operations emerged and attracted investment. Representative examples include AMP Robotics (founded 2015, Louisville, CO), which developed AI-powered material sortation systems for mixed-waste recovery facilities, and Greyparrot and Recycleye (both founded 2019, London), which offer AI waste analytics platforms and robotic optical sorting systems respectively. Industry data indicate that over 60 robotic waste-sorting start-ups are currently active globally, with an average of approximately four new entrants per year over the past decade. The G06V (image recognition), G06N (machine learning), H04W (wireless networking), and H04L (digital data transmission) subclass increases documented in this study are consistent with the technical building blocks underlying these commercial applications, suggesting that patent activity and market entry have concentrated in overlapping functional domains. Future research linking patent assignee data with firm-level databases would allow more systematic examination of the relationship between emerging patent clusters and SME formation trajectories across technology domains.

The inflection point of 2015 identified in this study also aligns closely with a wave of regulatory and policy signals that provided direct incentives for the types of innovation captured in the patent data. The European Union’s 2015 Circular Economy Action Plan articulated concrete transition tasks promoting circularity, building markets for secondary materials, and strengthening recyclability requirements. These policies provided regulatory incentives for investment in sorting infrastructure, material traceability, and process efficiency areas corresponding to the G06, H04, G01, and B65 growth clusters identified in this study. Subsequent directives reinforced and extended these signals: the revised 2020 Circular Economy Action Plan broadened the focus toward data-driven circular system implementation, the Single-Use Plastics (SUP) Directive established requirements that provided regulatory impetus for new collection and sorting solutions, and the Eco-design for Sustainable Products Regulation (ESPR) established product-level circularity requirements that necessitate traceability and standardization technologies. Beyond the EU, comparable regulatory developments in other major markets including extended producer responsibility schemes and national recycling targets contributed to a globally synchronized surge in recycling-oriented patent activity after 2015. The timing of the portfolio shifts observed in this study is consistent with these regulatory frameworks having directed R&D attention toward operational and digital domains rather than basic chemistry, though the causal relationship between regulatory signals and patent filing behavior cannot be established from the patent data alone.

The emerging technological directions identified in this study can be further grounded by connecting the observed patent portfolio shifts to specific technology do-mains reported in the prior literature [84]. Four areas are particularly illustrative. First, the post-2015 expansion of G06V (+0.23 pp), G06N (+0.57 pp), and G01N subclasses in this study corresponds to the domain of AI- and sensor-based plastic sorting and quality discrimination. Reviews of sensor-based solid waste sorting have documented applications of near-infrared, Raman, and laser-induced breakdown spectroscopy integrated with chemometric models for plastic waste classification [38], and hyperspectral sensing combined with machine learning algorithms for plastic detection and sorting [39], while systematic comparisons across sensor types and waste sorting contexts have identified data preprocessing, model robustness, and contamination handling as central technical challenges [37]. The structural growth of image recognition, computational modeling, and analytical inspection subclasses in the patent portfolio is consistent with R&D attention concentrating in precisely these areas. Second, the pronounced increases in H04W (+1.78 pp) and H04L (+1.01 pp), alongside G06F (+1.05 pp), correspond to the integration of IoT and digital connectivity into recycling operations. Reviews of AI- and IoT-enabled smart waste management have synthesized evidence on route optimization, sorting accuracy improvement, and operational efficiency applications [12,77]; the selective strengthening of wireless networking and digital data transmission subclasses identified in this study aligns with this direction, specifically in the functions required for real-time process data transmission and equipment interoperability across the recycling value chain. Third, the increases in C09K (+0.38 pp) and C09J (+0.13 pp), together with the topic transition from Polymer coating to Chemical recycling for packaging identified in the BERTopic analysis, correspond to the domain of advanced chemical recycling and feedstock quality recovery. The demonstration of engineered PET depolymerization enabling high-purity monomer recovery from post-consumer bottles [40] represents a process-level example consistent with the observed shift toward post-treatment, formulation, and quality correction sub-classes in the later-period portfolio. Fourth, the rise in C02F (+0.62 pp) across both the IPC portfolio analysis and the topic-IPC mapping in Table 3 corresponds to growing R&D attention toward process-water management and washing-related treatment requirements. Collectively, these correspondences between the patent-based structural shifts and technology directions reported in the prior literature help to ground the otherwise abstract portfolio trends in recognizable technical domains, while the patent evidence itself documents changes in R&D attention rather than outcomes of any specific application.

6. Conclusions

This study quantitatively examined the long-term evolution of plastic recycling technologies using 64,639 triadic patents filed between 2005 and 2024 and comparatively analyzed shifts in technological structure and thematic focus across the pre- and post-2015 periods, namely 2005–2014 and 2015–2024. The analysis combined IPC portfolio analysis, co-occurrence network analysis, and BERTopic-based topic modeling to characterize how technological portfolios and thematic structures were reconfigured across the two periods.

The results show that, at the section level, Sections C and B remained dominant foundational axes across the full period, while the post 2015 period exhibits a relative expansion of Section B and stronger coupling with elements associated with Sections H and G. At the class and subclass levels, increases are concentrated in G06, data processing and computing, H04, communications and connectivity, G01, measurement and inspection, B65, logistics and handling, C02F, water treatment, and C09, formulation and quality correction. These patterns suggest a shift in the composition of the patent portfolio, in which operations centered functions spanning sorting, washing, quality management, traceability, and operational optimization have increasingly expanded their share within the upper tier of the technology portfolio, as measured by IPC share and co-occurrence frequency. Co-occurrence networks further reveal a reconfiguration of convergence axes from early-period linkages centered on polymer formulation and synthesis toward later-period combinations that integrate processing, pre- and post-treatment, standardization, and elements related to logistics and environmental management. Topic-transition results similarly show that later-period themes more tightly coupled to circular value chains and operating systems become more prominent, including chemical recycling, particularly for packaging, reverse logistics for reusable packaging, remanufacturing, system recovery, and data sources.

These findings provide empirical evidence that patent activity in plastic recycling is not exclusively concentrated in individual conversion processes or material performance domains. Instead, the IPC portfolio and co-occurrence data show that patenting has increasingly extended into operational capabilities and data-driven integration functions spanning sorting, quality management, traceability, and process optimization. These patent-based patterns suggest the potential value of portfolio designs that combine process cores with operational infrastructure, as the domains that have grown most rapidly in share terms are those associated with sorting, connectivity, quality management, and logistics. From a policy perspective, the concentration of later-period patent growth in G06, H04, G01, C02F, and B65 suggests that support directed at sorting infrastructure, traceability and data standards, quality measurement and certification systems, and criteria for process-water and contamination management would be directed at the functional domains where patenting activity has demonstrably intensified.

Several limitations should be noted. First, as a general constraint of patent-based analysis, application strategies, examination regimes, and differences in appropriation practices may introduce biases such that patent counts do not fully represent the scale of underlying technological activity. Second, although IPC provides a systematic taxonomy of technologies, its resolution and the timing of labeling can lag early diffusion of emerging technologies, and similar technologies may be dispersed across multiple codes. Third, topic-modeling outputs include some adjacent-industry topics weakly linked to the recycling domain, reflecting the challenge of achieving perfect domain purity given the breadth of the search query and the characteristics of triadic patent data. Fourth, because the study explains structural change primarily through share shifts and co-occurrence-based convergence, it does not directly link these structural dynamics to performance indicators such as recycled-material properties, process costs, or carbon-mitigation effects.

Future research can address these limitations in several ways. First, beyond volume-based indicators, it is important to incorporate qualitative patent metrics such as forward and backward citations, patent-family expansion, and proxies for claim scope, including the number and length of claims, to evaluate technological influence and diffusion speed. Second, domain purity can be improved by combining IPC-based analysis with CPC, keyword-based technology dictionaries, or classification refinement using embedding- or Large Language Model (LLM)-based methods, alongside systematic procedures for reducing topic contamination. Third, co-occurrence networks can be extended beyond static comparisons by using dynamic network analyses based on annual snapshots and by tracking changes in centrality and community structures to identify the timing of convergence formation and potential transition thresholds. Finally, empirical scope can be broadened by decomposing and comparing portfolios by process route, such as mechanical, chemical, thermal, and biological pathways, by product group, such as packaging, textiles, automotive, and electrical and electronic products, or by region, and by quantitatively linking structural changes in patenting to exogenous variables such as policy shifts, oil prices, or the timing of regulatory adoption. These extensions would enable more precise testing of how the observed transition toward operations- and data-driven integration relates to industrial and policy conditions and how it translates into measurable outcomes in quality, economics, and carbon reduction. Beyond methodological extensions, future research could also pursue real-world validation through case studies that examine how the technology combinations identified as structurally prominent in this patent analysis—such as the integration of AI-based sorting, digital traceability, and quality correction—manifest in specific recycling facilities or value chains, and whether the patent-level structural shifts are reflected in operational or material quality outcomes at the firm or system level.