Strategic Management Knowledge Map via BERTopic (1980–2025): Evolution, Integration, and Application

Abstract

Problem: Amid digital disruption and the cross-fertilization of RBV, DCV, and KBV, strategic management knowledge has grown fragmented with blurred boundaries. Conventional mapping (citation/co-word, LDA) lacks semantic and temporal resolution, obscuring overlaps, divergences, and turning points and hindering links to actionable indicators (e.g., the Balanced Scorecard). Hence, an integrated, semantically faithful, time-stamped map is needed to bridge research and operational metrics. Gap: Prior maps rely on citation/co-word signals, miss textual meaning, and treat RBV/DCV/KBV in isolation—lacking a theory-aligned, time-stamped, manager-oriented synthesis. Objectives: This study aims to (1) reveal how RBV, DCV, and KBV evolve and interrelate over time; (2) produce an integrated, semantically grounded map; and (3) translate selected themes into actionable managerial indicators. Method: We analyzed 25,907 WoS articles (1980–2025) with BERTopic (Sentence-BERT + UMAP + HDBSCAN + c-TF-IDF). We used an RBV/DCV/KBV lexicon to guide retrieval/interpretation (not to constrain modeling). We discovered 230 topics, retained 33 via coherence (C_V), and benchmarked them against LDA. Key findings: A concise set of 33 high-quality themes with a higher C_V than LDA on this corpus was established. A Fish-Scale view (overlapping subfields across economics, management, sociology) that clarifies RBV–DCV–KBV intersections was achieved. Era-sliced prevalence shows how themes emerge and recombine over 1980–2025. Selected themes mapped to Balanced Scorecard (BSC) indicators linking capabilities → processes → customer outcomes → financial results. Contribution: A clear, time-aware synthesis of RBV–DCV–KBV and a scalable, reproducible pipeline for structuring fragmented theory landscapes are presented in this study—bridging scholarly integration with managerial application via BSC mapping.

1. Introduction

1.1. Research Background

In the context of rapid knowledge evolution and increasing cross-disciplinary integration, strategic management has become a highly integrative field marked by dynamic development and theoretical convergence. Its intellectual lineage runs from early management pioneers [1,2] through Porter’s external orientation [3,4] to internal resource views 2014 Resource-Based View (RBV) [5] and Dynamic Capabilities View (DCV) [6]—which emphasize sensing, learning, and reconfiguring capabilities in changeable environments.

In the digital era, strategy further shifts toward digital and ecosystem orientations [7,8], ecosystem perspectives [9,10], and synthesis of dynamic capabilities, platforms, and business models [11]. These developments broaden sources of insight across economics, sociology, psychology, and management and blur theoretical boundaries, leading to paradigm expansion, cross-theory integration, and overlap [12,13].

Following Campbell’s Fish-Scale Multiscience model, we view strategic management as a pluralistic knowledge system in which overlapping “scales” (Root Disciplines) provide complementary but partial coverage. Core theories draw simultaneously on economics (e.g., transaction cost economics [14]; RBV [5]), sociology (e.g., institutional theory [15]; resource dependence [16]), psychology (e.g., behavioral theory [17]; organizational learning [18]; upper echelons [19]), and management (e.g., dynamic capabilities [6]). Accordingly, our maps use Fish-Scale to make disciplinary provenance explicit (Root Discipline–Field–Subfield), visualize cross-root overlap via topic similarity networks, and provide time-sliced views of how configurations relate to shifting disciplinary influences. We do not pre-specify any evolutionary form in the Introduction Section; Section 3 details estimation, and Section 4 reports empirical patterns. Fish-Scale thus serves as a principled rationale for multi-root membership and overlap rather than a post hoc metaphor [12,13].

Despite a rich theoretical landscape, prior work on strategy’s evolution has relied mainly on co-citation analysis (e.g., [20], mapping SMJ 1980–2000; identifying influential works such as Porter and Strategy [3]; Barney [5]; Wernerfelt [21]). Such approaches, while foundational, (1) depend on citation frequency rather than semantic content, (2) emphasize bibliometric links over textual substance—under-detecting topic fusion and theoretical innovation—and (3) are dated relative to digital and AI-driven developments.

To address these issues, scholars increasingly use topic modeling on full text. Latent Dirichlet Allocation (LDA) [22] discovers topics without manual labels but, due to its bag-of-words assumption, under-captures context [23,24]. This motivates semantically richer approaches.

Our aim is to provide a theory-informed, semantically grounded, time-aware map of strategic management that clarifies relationships among major traditions and supports practical application in empirical industries. We implement this using a semantic topic modeling workflow and organize outputs with Fish-Scale for surface disciplinary provenance. Full methods and visualizations are presented in Section 3 and Section 4, with managerial translation shown in Section 5.

1.2. Brief Literature Context

A detailed account of external positioning, RBV/KBV/DCV developments, pluralism and fragmentation across innovation, ecosystems, and digital transformation, and bibliometric mappings is presented in Section 2 [4,5,6,20,25,26]. We summarize only essential context here to keep the Introduction Section concise.

1.3. Methodological Background

Technical background on Natural Language Processing (NLP), LDA [22], BERT/Sentence-BERT [27,28], and BERTopic [29]—including UMAP [30], HDBSCAN [31], and class-based TF-IDF—is provided in Section 3, along with coherence validation (C_V, C_NPMI [32] and configuration details. We avoid method specifics in the Introduction Section to maintain readability.

1.4. Objectives and Contributions

Objectives. This study pursues four objectives aligned with a theory-informed, semantically grounded, and time-aware mapping of strategic management:

• O1 (Topic discovery and quality): Identify latent topics in a domain-bounded corpus (RBV, KBV, DCV, CA) and retain high-quality topics based on coherence (C_V, C_NPMI [32]).
• O2 (Theoretical mapping): Map retained topics to the Fish-Scale hierarchy (Root Discipline → Field → Subfield) to make disciplinary provenance explicit [12,13].
• O3 (Temporal structure): Provide a time-sliced view of topic configurations to describe shifts in emphasis across strategic traditions (1980–2025).
• O4 (Methodological benchmarking): Benchmark the embedding-based approach (BERTopic [29]) against LDA [22] on coherence and interpretability.

Contributions. This study contributes via the following:

• C1 (Substantive): Delivering an interpretable, theory-aligned map that clarifies how RBV/KBV/DCV/CA relate and evolve across eras [4,5,6].
• C2 (Methodological): Demonstrating a semantics-first pipeline (Sentence-BERT → UMAP → HDBSCAN → c-TF-IDF) that outperforms LDA on coherence for dense theoretical text [28,29,31,32].
• C3 (Practical): Providing a reproducible basis for topic labeling, construct classification, and time-aware knowledge mapping useful to both scholars and managers, with a managerial translation via Balanced Scorecard [33].

The structure of this article is as follows: Section 2 reviews related work. Section 3 describes research methodology. Section 4 reports the 33-topic structure, disciplinary provenance (Fish-Scale hierarchy), and time-sliced visualizations. Section 5 discusses managerial implications and demonstrates a Balanced Scorecard (BSC) translation [33]. Section 6 concludes this paper. Note. A complete list of abbreviations in this study is provided in Appendix A.

2. Literature Review

2.1. Scientific Knowledge Evolution Analysis

Amidst the growing scholarly attention to the dynamic capabilities theory, researchers have increasingly focused not only on the evolution of its core concepts but also on the diffusion of knowledge and the trajectory of theoretical development [34]. To systematically capture the developmental path of this theory within the academic domain, Scientific Knowledge Evolution Analysis has emerged as a vital analytical approach. This form of analysis reveals patterns of knowledge dissemination and the structure of scholarly influence within a given field. It aids in identifying key theoretical contributions and developmental milestones while reflecting shifts in thematic focus and theoretical frameworks within academic communities [35]. Building on the above theoretical foundations, strategic management exemplifies a Fish-Scale Multiscience structure [13], in which overlapping yet specialized domains from economics (e.g., transaction cost economics, resource-based view), sociology (e.g., institutional theory, resource dependence theory), and psychology (e.g., organizational learning, behavioral theory) intersect. This epistemological pluralism has fostered a dynamic and evolving knowledge system, characterized by theoretical fusion, differentiation, and regeneration. Such structural interweaving not only reflects the field’s interdisciplinary heritage but also explains its capacity to adapt to emerging research challenges and integrate diverse explanatory logics.

Contemporary research methods for such analysis generally fall into two strategic categories [36,37]. The first involves citation-based network tracing through Main Path Analysis (MPA), which maps the backbone of knowledge development by identifying influential citation pathways [38]. The second approach combines bibliometric techniques or text mining with topic modeling to extract thematic structures and track their evolution over time. This topic evolution tracking method integrates topic modeling with temporal analysis to uncover dynamic changes in scholarly discourse [39].

2.2. Knowledge Structure

2.2.1. Main Path Analysis in Citation Networks

Citation data have long been regarded as a key proxy for tracing the flow and diffusion of scientific knowledge [34]. Building on Garfield’s concept of historiography maps, new research ideas and discoveries are often constructed upon prior scholarly achievements. Grounded in this principle, Main Path Analysis (MPA) identifies the most influential trajectories of knowledge flow by analyzing the structural connections within citation networks, thereby revealing the evolutionary path of scientific development [40,41].

For instance, Huang, et al. [42] employed both global and local MPA to identify technological trends in dye-sensitized solar cells. Chen et al. [43] introduced a semantically enhanced MPA approach that successfully extracted multiple parallel technological trajectories within the lithium battery domain. Park and Magee [44] further developed a knowledge persistence-based method using forward and backward search strategies to pinpoint high-impact developmental trajectories based on patent persistence.

MPA emphasizes extracting representative knowledge development paths from citation relationships. By constructing citation networks and quantifying link strengths between nodes, researchers can clarify the most influential theoretical sources and developmental axes at different temporal stages. This method has been widely applied in high-tech domains and offers a powerful means for delineating knowledge evolution, identifying seminal contributions, and guiding research agendas. In strategic management—especially in dynamic capabilities research—MPA facilitates understanding of knowledge flow and theoretical transformation across periods, contributing to strategic positioning and theoretical consolidation.

2.2.2. Topic Extraction and the Mechanism of Evolutionary Pathways

Topic identification can also be achieved through keyword co-occurrence, co-citation, or bibliographic coupling networks, which reveal the clustering and structural associations among knowledge units [45,46]. Visualization tools and community detection algorithms further support the analysis of topic transitions and the evolution of knowledge over time [47].

Empirical studies affirm the effectiveness of these methods. Katsurai and Ono [48] used co-word networks to track topic shifts across disciplines. Majdouline, et al. [49] employed network indicators to quantify temporal patterns in co-citation structures in technology entrepreneurship, while Mariani and Borghi [25] integrated bibliographic coupling and network analysis to explore Industry 4.0’s conceptual development.

Text vectorization techniques such as TF-IDF and Latent Dirichlet Allocation (LDA) are widely used in topic modeling [22]. For example, Yang, et al. [50] applied the Lingo algorithm to examine technological linkages, Debao, et al. [51] combined TF-IDF and K-means to map big data research themes, and Song and Suh [26] used topic modeling and network analysis to understand industrial safety innovation.

However, traditional models face limitations. TF-IDF lacks the capacity to represent semantic context Devlin, Chang, Lee and Toutanova [27]), and LDA struggles to model word-level semantics and syntactic relationships [52]. Both are constrained by the bag-of-words assumption, leading to sparse vectors and reduced efficiency in large datasets. To address these shortcomings, embedding-based methods are increasingly adopted to enhance semantic richness and computational scalability in analyzing topic evolution.

This study focuses on semantic topic extraction to uncover the latent thematic structure in strategic management research. While the present analysis centers on topic identification, the exploration of evolutionary pathways between topics will be a central objective in future research.

2.2.3. Applications and Advantages of BERT and BERTopic in Topic Extraction

BERT (Bidirectional Encoder Representations from Transformers) is a deep learning model that generates context-sensitive embeddings through bidirectional training on text sequences [27]. Unlike static embedding models such as word2vec or doc2vec, BERT captures nuanced semantic relationships and is well suited for identifying latent themes in complex academic texts [27].

Domain-specific variants such as SciBERT [53] and BioBERT [54] improve performance by training on scientific and biomedical corpora, respectively.

BERTopic builds upon BERT embeddings, applying UMAP for dimensionality reduction, HDBSCAN for clustering, and class-based TF-IDF (c-TF-IDF) for keyword extraction [29]. It offers advantages over LDA by avoiding the need to predefine the number of topics and by capturing richer semantic structures. The model’s strength lies in its ability to identify interpretable and temporally dynamic topic clusters across large, unstructured corpora.

Its application in various domains has been validated: Moreno, et al. [55] explored Airbnb user preferences through semantic clusters; Khodeir and Elghannam [56] analyzed learner feedback in MOOCs. These examples highlight the model’s potential for trend detection, especially in multilingual and noisy textual datasets.

Nonetheless, BERT-based models have high computational demands, and topic labeling remains a challenge due to the abstract nature of cluster contents. Further development in automatic topic naming and interpretability metrics will enhance the utility of these methods.

2.2.4. Textual Data Collection and BERT-Based Topic Modeling in Strategic Management

Strategic management research, housed in the Web of Science (WoS) database, spans foundational theories such as the Resource-Based View (RBV), Industrial Organization (IO), and Transaction Cost Economics (TCE), as well as emerging areas like Dynamic Capabilities, Platform Strategy, and Digital Transformation. These texts are often abstract, theoretically dense, and conceptually overlapping, making them difficult to analyze with frequency-based methods like TF-IDF or LDA [27].

To address these challenges, this study applies BERTopic to model the semantic structure of strategic management research. By leveraging Sentence-BERT embeddings, UMAP, and HDBSCAN, the method detects latent topics and captures temporal dynamics without relying on predefined topic counts [57]. Applied to WoS research from 1980 to 2025, BERTopic identifies themes such as resource orchestration, environmental dynamism, open innovation, and digital platform strategy, enabling visualization of topic trends and intertopic relationships. Recent advances show that transformer-based knowledge embeddings substantially improve domain-specific semantic understanding and decision quality. For instance, Zhu, et al. [58] demonstrate a knowledge-embedding transformer for medical QA that outperforms GPT-4 on MedQA by leveraging a retrieval/understanding layer before generation; likewise, Khan, et al. [59] propose a Joint Multi-Scale Multimodal Transformer (JMMT) that captures inter- and intramodal relationships across visual and audio signals, highlighting how multi-scale attention enhances semantic fusion. While our task is topic modeling rather than QA or multimodal classification, these works collectively support our choice of an embedding-centric, semantics-first pipeline (sentence embeddings + c-TF-IDF) and our focus on coherence (C_V, C_NPMI) as proxies of semantic integrity in theory-driven mapping.

To enhance accuracy, domain-specific models such as SciBERT [53] and PatentSBERTa are utilized. These models improve alignment with scholarly discourse and help uncover theoretical structures in complex research. Coherence metrics such as C_V and NPMI are employed to validate the interpretability of extracted topics.

In summary, BERT and BERTopic offer a novel methodological advancement for mining and analyzing strategic management research. These tools overcome key limitations of traditional models by enabling deeper semantic understanding and more accurate extraction of research themes. They also provide a strong foundation for building strategic theory knowledge maps, conducting evolutionary analysis, and exploring theoretical convergence across disciplines.

3. Research Methodology

This study employs an integrated topic modeling framework to map the semantic structure and temporal evolution of strategic management research. Figure 1 summarizes four stages: (1) metadata completeness screening—records lacking any of title, abstract, or author keywords (required to construct the composite analytical field, text_all) were removed; (2) sentence-level semantic embedding (BERT/Sentence-BERT); (3) dimensionality reduction via UMAP; and (4) density-based topic clustering with HDBSCAN. The workflow is implemented in BERTopic, which integrates these steps and uses class-based TF-IDF to extract interpretable topic keywords.

Initial retrieval yielded 31,639 records from Web of Science (1980–2025). After excluding the documents with none of the information from the data file columns, (title, keywords, and abstract), a curated corpus of 25,907 documents was retained for modeling; all BERTopic analyses reported below were conducted on this filtered corpus. BERTopic was then applied to extract latent topic structures and calculate coherence scores, allowing for the selection of 33 high-quality topics with coherence values exceeding 0.8.

Compared to conventional models such as LDA, BERTopic offers improved semantic sensitivity and topic interpretability by leveraging transformer-based embeddings. The final topics were subsequently aligned with a Root Discipline–Field–Subfield classification scheme to map their theoretical positioning within the broader landscape of strategic management.

The research process is structured into three major stages: data collection, data preprocessing, and data analysis, as illustrated in Figure 2. First, 31,639 documents related to strategic management were retrieved from the Web of Science (WoS) database. A refined corpus of 25,907 documents was filtered for analysis. In the preprocessing stage, textual normalization, stopword removal, and n-gram construction were performed to enhance the capture of strategic semantics.

Subsequently, semantic embeddings and clustering were conducted using the BERTopic model. Topic coherence was assessed through both the C_V and C_NPMI metrics to ensure internal consistency. Topic classification was further refined using a theory-guided framework by aligning topics with established Root Discipline–Field–Subfield taxonomies.

Finally, the performance of the BERTopic model was compared with traditional LDA-based topic modeling to evaluate coherence and interpretability across methodological approaches.

3.1. Data Collection and Literature Screening

The data for this study were retrieved from the Web of Science (WoS) Core Collection, a multidisciplinary citation database renowned for its rigorous indexing and coverage of high-quality journals. To ensure consistency and comparability, the analysis focused on journal articles published in English within the domain of strategic management from 1980 to 2025.

We conducted a title-field (TI) search targeting core strategic management theories—Resource-Based View (RBV), Dynamic Capabilities View (DCV), Knowledge-Based View (KBV), and Competitive Advantage (CA)—using the representative phrases (e.g., “resource-based view,” “dynamic capabilities,” “knowledge management,” “competitive advantage”) shown in

Table 1. Search strategy.

Constructs	Core Theory	Keywords
Knowledge-Based View (KBV)	Knowledge-Based View	knowledge management, knowledge sharing, knowledge transfer, knowledge acquisition, knowledge integration, knowledge exploitation, knowledge exploration, knowledge capital
	Intellectual Capital Theory	intellectual capital, human capital, structural capital, relational capital, social capital
	Organizational Learning Theory	organizational learning, learning capability, absorptive capacity, learning organization
	Knowledge-Based Innovation	knowledge-based innovation, knowledge integration capability, knowledge transformation, knowledge absorption, knowledge reconfiguration
Resource-Based View (RBV)	Resource-Based View (RBV)	resource-based view, resource-based theory, resource advantage theory, firm resources, tangible resources, intangible resources, core competencies
	VRIO Framework	VRIO framework, value, rarity, imitability, organization, valuable resources, rare resources, imitable resources, non-substitutable resources
	Strategic Resource Theory	strategic resources, competitive resources, strategic assets, firm-specific resources
	Resource Typology	human resources, organizational resources, technological resources, financial resources
Dynamic Capabilities View (DCV)	Dynamic Capabilities View (DCV)	dynamic capabilities, DCV, dynamic capability view, sensing, seizing, transforming, strategic flexibility, organizational agility, adaptive capabilities, market responsiveness
	Resource Reconfiguration Theory	resource reconfiguration, technological adaptation, knowledge reconfiguration
	Innovation & Agility	innovation capability, learning agility, competitive agility, strategic adaptation, business model innovation
Competitive Advantage (CA)	Competitive Advantage Theories	competitive advantage, sustainable competitive advantage, long-term competitive advantage, strategic positioning, cost leadership, differentiation strategy, focus strategy
	Value Creation Theories	value creation, value co-creation, customer value
	Performance-Based Perspective	firm performance, strategic performance, innovation performance, financial performance, firm growth, business performance
	Market and Strategic Positioning	market competition, competitive positioning

. The results were restricted to 1980–2025, document type of “article”, and Web of Science subject categories in Business, Management, Economics, and Business Strategy, yielding 31,639 records; then, following PRISMA (Identification → Screening → Eligibility → Inclusion), the predefined search identified 31,639 records. During screening, records lacking the metadata required to construct the composite analytical text field (title, abstract, or author keywords) were removed. At eligibility/inclusion, 25,907 records were retained and carried forward to modeling. Thus, the reduction from 31,639 to 25,907 reflects metadata completeness and routine hygiene rather than any theory filter.

TI = ((“knowledge management” OR “knowledge sharing” OR “knowledge transfer” OR “knowledge acquisition” OR “knowledge integration” OR “knowledge exploitation” OR “knowledge exploration” OR “intellectual capital” OR “knowledge capital” OR “human capital” OR “structural capital” OR “relational capital” OR “social capital” OR “organizational learning” OR “learning capability” OR “absorptive capacity” OR “knowledge-based innovation” OR “knowledge-based view” OR “learning organization”) OR (“resource-based view” OR “RBV” OR “resource-based theory” OR “resource advantage theory” OR “firm resources” OR “tangible resources” OR “intangible resources” OR “human resources” OR “organizational resources” OR “technological resources” OR “financial resources” OR “core competencies” OR “strategic resources” OR “valuable resources” OR “rare resources” OR “imitable resources” OR “non-substitutable resources” OR “VRIO framework” OR “value, rarity, imitability, organization” OR “resource orchestration” OR “firm-specific resources” OR “competitive resources” OR “strategic assets” OR “dynamic resources”) OR (“dynamic capabilities” OR “DCV” OR “dynamic capability view” OR “sensing, seizing, transforming” OR “strategic flexibility” OR “organizational agility” OR “adaptive capabilities” OR “market responsiveness” OR “technological adaptation” OR “resource reconfiguration” OR “innovation capability” OR “learning agility” OR “competitive agility” OR “strategic adaptation” OR “business model innovation” OR “knowledge reconfiguration” OR “knowledge absorption” OR “knowledge transformation” OR “knowledge integration capability”) OR (“competitive advantage” OR “sustainable competitive advantage” OR “long-term competitive advantage” OR “competitive positioning” OR “cost leadership” OR “differentiation strategy” OR “focus strategy” OR “strategic positioning” OR “market competition” OR “value creation” OR “value co-creation” OR “customer value” OR “firm performance” OR “strategic performance” OR “competitive positioning” OR “innovation performance” OR “financial performance” OR “firm growth” OR “business performance”)) AND PY = (1980–2025) AND WC = (“Business” OR “Management” OR “Economics” OR “Business Strategy”) AND DT = (“Article”).

Although the initial corpus was constructed using a theory-driven search strategy, the application of BERTopic provides a deeper, inductive, and semantically grounded classification of the strategic management literature. It enables the identification of latent topic structures, inter-theory dynamics, and temporal evolution patterns that static keyword filters cannot reveal, thereby advancing both methodological rigor and theoretical clarity.

3.2. Data Preprocessing

Construction of Text Corpus

To prepare for topic modeling and semantic analysis, a structured text dataset was compiled from the Web of Science (WoS) Core Collection. The dataset includes four essential metadata fields, title, abstract, author keywords, and year, stored in .xlsx format for efficient processing and programmatic integration.

Records lacking abstracts or English keywords were excluded to ensure semantic completeness. Each article was represented as a structured entry $T_i$, and the full dataset can be formalized as follows:

$$D = \{T_i\mid i\le n\}$$

(1)

where $n$ is the total number of articles, and each $T_i$ contains the four aforementioned fields. This dataset serves as the input for subsequent BERT-based topic modeling and semantic coherence evaluation.

3.3. Semantic Topic Modeling with BERTopic

To investigate the semantic structure and topic evolution in the strategic management literature, BERTopic was adopted as the core modeling framework. Compared to traditional methods such as Latent Semantic Analysis (LSA) or Latent Dirichlet Allocation (LDA), BERTopic combines semantic embeddings and density-based clustering (e.g., HDBSCAN), providing enhanced accuracy in capturing latent topic structures within large-scale unstructured text.

In this study, we applied BERTopic to the filtered corpus of 25,907 WoS articles. To improve representational richness, we used Sentence-BERT for semantic embedding and applied trigram-based CountVectorizer to preserve key phrase structures. The modeling procedure was optimized for topic stability, coherence evaluation, and interpretability through visualization.

3.3.1. BERTopic Topic Modeling

In this study, we employed abstracts from 31,639 strategic management articles retrieved from the Web of Science (WoS) Core Collection as the primary corpus for topic modeling. To ensure the comprehensiveness and interpretability of semantic representation, we utilized the Sentence-BERT model for semantic embedding. Additionally, a trigram-based CountVectorizer was applied to preserve key multi-word expressions during vectorization.

This configuration was designed to enhance the stability of the number of extracted topics and facilitate subsequent evaluations of topic coherence and visual interpretability. The algorithmic procedure (in 3 Algorithm 1) used in this study is detailed as follows:

Algorithm 1. BERTopic-based topic modeling.

Corpus D (Title/Abstract/Keywords composites; n = 25,907). Require: Corpus D; parameters embedding_model, vectorizer Ensure: Topics T, probabilities P, keyword sets K 1: Normalize text; tokenize; remove stopwords. 2: Embed with Sentence-BERT. 3: Reduce with UMAP; cluster with HDBSCAN. 4: Extract keywords via c-TF–IDF; compute C_V, C_NPMI. 5: return (T, P, K)

BERTopic initially generated 230 topics. To improve interpretability, we applied the optional reduce_topics function to merge semantically overlapping clusters, resulting in 33 high-coherence topics (C_V > 0.8). This reduction step was applied only after the model had discovered topics in an unsupervised manner, ensuring consistency with BERTopic’s capability to infer topic structures without predefining their number.

3.3.2. Topic Coherence Evaluation

To validate the semantic integrity of the modeled topics, this study employed topic coherence as a quantitative evaluation metric. As suggested by [32], the coherence score effectively reflects the semantic relatedness among keywords within a topic, making it a widely recognized standard for assessing topic modeling quality.

Specifically, after generating topics using the BERTopic model, we extracted the top 10 representative keywords for each topic and used the C_V coherence metric as our evaluation standard. The C_V score considers both the co-occurrence context and semantic similarity, providing a stable and interpretable basis for assessing coherence in multi-topic environments.

The following procedure (in Algorithm 2) outlines the coherence score calculation process used in this study:

Algorithm 2. Topic score calculation.

Corpus D (Title/Abstract/Keywords composites; n = 25,907). Input: Structured dataset from topic-modeling results (full text or tokenized) Output: Coherence score C_V for topic coherence Extract topic words and keywords 1: topic_words ← GETTOPICSANDWORDS() Prepare text format for coherence calculation 2: (texts, dictionary, corpus) ← PREPARECOHERENCEDATA(df) Set number of keywords per topic 3: topn_words ← 10 Generate topic–term sets 4: topic_terms ← GETTOPICTERMS(topic_words, topn_words) Calculate coherence score 5: coherence_score ← EVALUATECOHERENCESCORE(topic_terms, texts, dictionary) 6: return coherence_score       ▷ C_V

Modeling followed the main pipeline—Sentence-BERT embeddings, UMAP reduction, HDBSCAN clustering, and c-TF-IDF keywording—implemented via BERTopic, including topic naming and coherence evaluation (C_V, C_NPMI). This establishes a robust, reproducible text mining workflow that balances theoretical grounding with semantic precision. The approach reveals a latent thematic structure in strategic management and competitive advantage, traces temporal distributions, and situates cross-theory linkages. It overcomes limits of bag-of-words models and supports defensible topic labeling, construct classification, and knowledge structure visualization.

3.4. Theory-Informed Knowledge Map: Definition and Assembly

The knowledge map is conceived as a theory-aligned, topic-level representation of the field that jointly depicts topic structure, relations among topics, disciplinary provenance, and their temporal configuration. Building on the unsupervised BERTopic results described in Section 3.1, Section 3.2 and Section 3.3, we assemble the map by first consolidating semantically redundant clusters with the optional reduce_topics routine, yielding an interpretable solution of 33 topics. Each topic is then assigned a concise, theory-recognizable label and aligned to a Root Discipline–Field–Subfield taxonomy to encode disciplinary provenance. Inter-topic semantic similarity is computed to render a network in which node size reflects prevalence, node color indicates provenance, and edge width encodes proximity; this view captures the structural layout of the domain. To incorporate dynamics, document–topic assignments are aggregated by publication era and visualized as a time-sliced flow, revealing how the configuration of topics changes across periods. Finally, to demonstrate managerial applicability, one representative topic is translated into a Balanced Scorecard (BSC) strategy map that articulates objectives across the Financial, Customer, Internal-Process, and Learning-and-Growth perspectives.

3.5. Topic Labels and Disciplinary Provenance

Topic labels are derived from c-TF-IDF keywords and exemplar documents and are reviewed to ensure brevity and distinctiveness. Provenance is determined by mapping salient terms to a controlled vocabulary spanning economics, sociology, psychology, and management; assignments are verified by two reviewers, and multi-root designations are permitted when the evidence indicates overlapping disciplinary signals. These provenance tags are used consistently in the visualizations (e.g., node coloring in the network) and in subsequent interpretation.

4. Result Analysis

This section presents the results of our topic modeling analysis on strategic management and competitive advantage research. Following the methodology in Section 3, we used semantic embedding and BERTopic to identify coherent and interpretable topics across a large corpus. We report the topic structures, representative publications, topic labels, and coherence scores, linking them to core theoretical frameworks. The analysis reveals the latent themes and semantic patterns within the literature, providing a foundation for theoretical integration and future application.

4.1. Topic Modeling Results Based on BERTopic

This study applied BERTopic to analyze 25,907 abstracts related to strategic management and competitive advantage, filtered from an initial retrieval of 31,639 records related to strategic management and competitive advantage, revealing a structured and semantically rich set of topics.

Figure 3 presents the top keyword distributions across 10 core topics, while

Table 2. Topic summary: keywords and provisional labels.

Topic ID	Representative Keywords (Top 5)	Provisional Topic Label
Topic 0	intellectual capital, capital, structural capital	Intellectual Capital and Knowledge Assets
Topic 1	family, family firms, family business	Family Business and Organizational Capital
Topic 2	km, knowledge management, management km	Knowledge Management and Resource Configuration
Topic 3	dynamic capabilities, capabilities, sic	Dynamic Capabilities and Market Adaptation
Topic 4	corporate social responsibility (CSR), corporate social responsibility, responsibility	Corporate Social Responsibility and Sustainable Advantage
Topic 5	absorptive, absorptive capacity, innovation	Absorptive Capacity and Innovation Potential
Topic 6	knowledge sharing, trust, employees	Knowledge Sharing and Organizational Trust
Topic 7	big data, analytics, data analytics	Big Data and Analytics-Driven Strategy
Topic 8	knowledge transfer, subsidiaries, multinational	Knowledge Transfer and Multinational Coordination
Topic 9	digital transformation, transformation, innovation	Digital Transformation and Innovation Governance

provides a summary of representative keywords and tentative topic labels derived from expert interpretation. In Figure 3, each bar indicates the relative importance of a keyword within its assigned topic. The longer the bar, the more representative or distinctive the keyword. The numerical values on the x-axis reflect c-TF-IDF scores, which capture how uniquely and frequently a term appears in a given topic relative to all others.

Each topic reflects a specific research domain within the broader field:

• Topic 0 emphasizes intellectual capital and knowledge assets, highlighting concepts such as “intellectual capital” and “structural capital,” central to knowledge-based views of competitive advantage.
• Topic 1 focuses on family business and organizational capital, characterized by terms like “family firms” and “family business,” suggesting a stream of research centered on governance, succession, and performance in family-owned firms.
• Topic 2 addresses knowledge management and resource configuration, incorporating phrases like “knowledge management” and “management KM,” reflecting discussions around dynamic knowledge capabilities and organizational learning.
• Topic 3 covers dynamic capabilities and market responsiveness, with strong associations with terms like “dynamic capabilities” and “capabilities,” aligned with the dynamic capabilities framework in the strategy literature.
• Topic 4 relates to corporate social responsibility and sustainable advantage, evidenced by terms such as “CSR” and “corporate responsibility,” capturing the intersection of ethics, legitimacy, and long-term performance.
• Topic 5 centers on absorptive capacity and innovation potential, including keywords like “absorptive capacity” and “innovation,” indicating a focus on external knowledge assimilation and innovation performance.
• Topic 6 highlights knowledge sharing and organizational trust, drawing on terms like “trust” and “employees,” and pointing toward collaboration, knowledge diffusion, and intraorganizational dynamics.
• Topic 7 involves big data and analytics-driven decision-making, showing strong signals of “data analytics” and “big data,” aligning with the rise of data-driven strategies.
• Topic 8 reflects knowledge transfer and multinational coordination, with terms such as “subsidiaries” and “multinational,” revealing insights into international strategy and global knowledge flow.
• Topic 9 focuses on digital transformation and innovation governance, combining terms like “digital transformation” and “innovation” to capture the evolution of strategic responses in the digital era.

These themes illustrate how the field bridges classical resource-based approaches with contemporary concerns such as digital disruption, sustainability, and globalization. The topic structure supports theory-driven naming and classification, enabling further mapping to strategic theories and trend evolution.

4.2. Topic Distribution and Topic Coherence Evaluation (BERTopic Modeling vs. LDA Topic Modeling)

This study evaluates the performance of BERTopic in identifying semantically coherent themes across a curated corpus of 25,907 scholarly documents, filtered from an initial dataset of 31,639 articles retrieved from the Web of Science. The objective is to uncover meaningful topic structures aligned with strategic management theories and trace their evolution from 1980 to 2025.

The BERTopic modeling process initially produced 230 distinct topics. To ensure thematic quality and theoretical relevance, topic coherence was assessed using two established metrics from the Gensim library: C_V, which reflects the semantic similarity among keywords within a topic, and C_NPMI, which measures normalized keyword co-occurrence. The overall model achieved a C_V score of 0.6620—surpassing the widely accepted threshold of 0.5—and a positive C_NPMI score of 0.1595, confirming that the topics are both semantically meaningful and statistically robust.

Based on these evaluations, 33 high-quality topics with C_V scores above 0.80 (top 85th percentile) were selected for further analysis. Each topic was labeled using a theory-driven lexicon and examined for conceptual significance and temporal dynamics. These topics exhibit strong coherence, interpretability, and alignment with foundational and emerging themes in strategic management.

To benchmark BERTopic’s performance, a comparative analysis was conducted using a Latent Dirichlet Allocation (LDA) model trained on the same dataset. The LDA model yielded a substantially lower C_V score of 0.4359, indicating weaker topic cohesion and interpretability (

Table 3. Topic coherence score comparison.

Method	C_V Coherence Score	C_NPMI Score	Interpretation
BERTopic	0.6620	0.1595	High topic coherence; topics are semantically consistent and interpretable
LDA	0.4359	-	Moderate coherence; less semantic alignment among topic terms

). This comparison underscores BERTopic’s advantage in capturing nuanced, overlapping themes in complex academic corpora through dynamic embeddings and density-based clustering. To complement the C_V and C_NPMI comparison, we provide three concrete examples where BERTopic yielded finer-grained, semantically coherent topics that LDA tended to merge in

Table 4. Illustrative examples (BERTopic vs. LDA).

Thematic Area	BERTopic (Separated)	LDA (Merged)	Why It Matters
Agency Theory Subdomains	Board Governance (e.g., board, directors, CEO, governance) and Executive Compensation (e.g., compensation, incentives, CEO pay)	One broad “governance/compensation” topic	Distinguishes monitoring vs. incentive mechanisms
ESG/Stakeholder vs. Legitimacy/Disclosure	Stakeholder/ESG (e.g., ESG, governance, performance) and Legitimacy/IC Disclosure (e.g., disclosure, reporting, intellectual capital)	Single CSR/ESG/disclosure topic	Separates stakeholder salience from legitimacy signaling

.

The finalized list of 33 coherent topics is presented in

Table 5. Thirty-three high-coherence topics.

Topic Id	Keywords	Topic Name
2	family, family firms, family firm, family business, ownership, family businesses, family involvement…	Socioemotional Wealth (SEW)
4	csr, corporate social, responsibility, social responsibility, corporate, social, responsibility csr…	Behavioral Agency Theory
13	hr, hrm, human resource, human, resource management, practices, resource, human resources, high perf…	Behavioral Theory of the Firm (HR, HRM, resource management, HPWS, human capital)
14	intellectual, intellectual capital, ic, capital, vaic, efficiency, capital efficiency, value added…	Resource-Based View (Intellectual Capital Efficiency)
16	board, directors, boards, ceo, corporate governance, governance, duality, director, busy, firm perfo…	Agency Theory (Board Governance)
15	gender, gender diversity, diversity, board, women, board gender, directors, female, boards, female d…	Behavioral Theory of the Firm (Board Gender Diversity)
19	compensation, pay, executive, ceo, executive compensation, ceo compensation, incentives, ceo pay, co…	Agency Theory (Executive Compensation)
23	corporate governance, governance, corporate, board, ownership, cg, firm performance, governance firm…	Agency Theory (corporate governance, board, ownership, agency)
25	disclosure, capital disclosure, icd, intellectual capital, intellectual, ic, reporting, capital, dis…	Legitimacy Theory
34	alliance, alliances, portfolio, alliance portfolio, partner, partners, portfolios, diversity, strate…	Network Theory
37	ceo, tmt, narcissism, ceos, leadership, upper, personality, ceo narcissism, upper echelons, echelons	Upper Echelons Theory (ceo, narcissism, upper echelons, leadership)
48	subsidiary, subsidiaries, reverse knowledge, reverse, transfer, knowledge transfer, rkt, multination…	Knowledge-Based View (KBV)
51	resilience, covid 19, covid, 19, pandemic, organizational resilience, 19 pandemic, crisis, disaster…	Dynamic Capabilities Theory (Crisis Resilience)
54	acquisitions, mergers, mergers acquisitions, border, cross border, merger, acquirers, acquirer, targ…	TCE (Cross-Border M&A Strategy)
55	ownership, shareholders, institutional, institutional investors, investors, governance, corporate go…	Agency Theory (ownership, investors, governance, performance)
58	diversity, tmt, diversity management, diversity firm, racial, gender, team, cultural diversity, gend…	Upper Echelons Theory (diversity, TMT, gender/cultural/age diversity)
61	esg, disclosure, environmental social, esg disclosure, social governance, governance esg, financial…	Stakeholder Theory
67	political, political connections, connections, lobbying, connected, politically connected, political…	Institutional Theory (Political Ties Strategy)
70	green, green supply, gscm, supply chain, supply, chain, chain management, environmental, gscm practi…	Natural Resource-Based View (Green Supply Chain)
74	islamic, islamic banks, banks, shariah, sharia, ibs, performance islamic, conventional, conventional…	Institutional Theory (Islamic Finance)
75	venture, venture capital, vc, cvc, corporate venture, backed, investors, vc backed, syndication, vcs	Resource-Based View + Real Options Theory (venture, VC)
109	exploration, exploitation, ambidexterity, exploration exploitation, exploitation exploration, ambide…	Ambidexterity Theory
117	franchising, franchise, franchisees, franchisor, franchisee, franchisors, transfer mechanisms, trans…	Agency Theory (franchising + knowledge transfer)
127	slack, slack resources, organizational slack, financial slack, slack firm, slack performance, hr sla…	Behavioral Theory of the Firm (slack, organizational)
124	succession, ceo, ceo succession, interim, founder, ceos, successor, turnover, successions, ceo turno…	Upper Echelons Theory (CEO succession, turnover, founder, interim)
132	audit, audit committee, committee, internal audit, ac, audit quality, committee characteristics, ac…	Agency Theory (audit committee, internal audit, ac effectiveness, audit quality)
138	ijvs, ijv, international joint, joint, joint ventures, ventures, international, foreign, acquisition…	Transaction Cost Economics (TCE) + Resource-Based View (RBV)
136	erp, resource planning, enterprise resource, implementation, erp implementation, planning erp, plann…	Dynamic Capabilities Theory (ERP Implementation)
155	acquisitions, acquirer, takeover, acquirers, bidders, takeovers, acquiring, mergers, returns, deals	TCE (M&A Performance and Valuation)
163	food, sscm, supply, supply chain, sustainable supply, chain, sustainable, food supply, chain managem…	Natural Resource-Based View (Sustainable Supply Chain)
183	inventory, inventory management, leanness, relationship inventory, inventory leanness, turns matrix…	Resource-Based View (Inventory Leanness Strategy)
184	iso, certification, 9000, iso 9000, 9000 certification, certified, iso 9001, 9001, standards, 14,001	Institutional Theory (ISO Legitimacy Strategy)
207	acquisitions, technological, acquiring, innovation performance, acquiring firms, cross border, borde…	Real Options Theory + Resource-Based View (RBV)

Note. “Topic ID” denotes the numeric identifier automatically assigned by BERTopic to each topic after the initial unsupervised discovery and the optional topic reduction step. IDs may be non-consecutive and are provided solely for cross-referencing; they do not indicate rank, importance, or chronology.

and forms the analytical basis for subsequent theory mapping and evolutionary analysis.

4.3. Theoretical Hierarchy of Strategic Management

To clarify the theoretical structure of strategic management, this study adopts the Fish-Scale Multiscience framework [12], which maps knowledge from Root Disciplines to Fields and Subfields, capturing overlaps across disciplinary boundaries.

Figure 4 presents a hierarchical visualization of the theoretical foundations underpinning strategic management research. At the innermost level, the diagram identifies four Root Disciplines—economics, sociology, psychology, and management—which serve as the epistemological bases of strategic thinking. Extending outward, these Root Disciplines give rise to intermediate fields, such as Contract Theory, Economic Sociology, Cognitive Psychology, and Business Ethics.

At the Subfield level, more specialized theories emerge, including Agency Theory, Behavioral Agency Theory, Upper Echelons Theory, Stakeholder Theory, and Network Theory. These perspectives provide focused analytical lenses for understanding firm behavior, governance structures, leadership dynamics, and stakeholder interactions (e.g., [19,60,61]).

This concentric structure clarifies the disciplinary lineage of the 33 BERTopic-derived topics, supporting their theoretical classification and mapping across the four strategic eras defined in this study.

4.4. Knowledge Map and Per-Topic Metrics

The reduced BERTopic solution yields 33 topics (Table 5). The topic network (Figure 5) shows a dense core organized around governance–behavioral and foundational strategy streams—most visibly connections among Agency Theory–Board Governance, Upper Echelons Theory (Board Gender Diversity), and Knowledge-Based View, with RBV/TCE variants linking nearby. Several smaller, specialized streams remain peripheral, indicating topical differentiation alongside the core.

The time-sliced view (Figure 6) indicates limited footprint before the 2000s, followed by sharp expansion in the 2010s–2020s. The three most prevalent topics—Socioemotional Wealth (SEW), Behavioral Agency Theory, and Behavioral Theory of the Firm (HR/HRM/HPWS/human capital)—account for a substantial share of that growth.

Table 6. Prevalence and coherence summary for 33 retained topics.

Topic_ID	Topic_Name	Size_Docs	Prevalence_pct	C_V	C_NPMI
2	Socioemotional Wealth (SEW)	367	1.42	0.8543	−0.0957
4	Behavioral Agency Theory	336	1.3	0.8673	0.2476
13	Behavioral Theory of the Firm (HR, HRM, resource management, HPWS, human capital)	182	0.7	0.8176	−0.0523
14	Resource-Based View (Intellectual Capital)	173	0.67	0.8657	0.1039
15	Upper Echelons Theory (Board Gender Diversity)	172	0.66	0.9628	0.2071
16	Agency Theory–Board Governance	168	0.65	0.9151	−0.1847
19	Agency Theory–Executive Compensation	151	0.58	0.8753	−0.0783
23	Agency Theory (corporate governance, board, ownership, agency)	133	0.51	0.8788	−0.2238
25	Legitimacy Theory	123	0.47	0.8555	−0.0486
34	Network Theory	102	0.39	0.8328	−0.1468
37	Upper Echelons Theory (ceo, narcissism, upper echelons, leadership)	98	0.38	0.8433	−0.2219
48	Knowledge-Based View	71	0.27	0.9215	0.0797
51	Dynamic Capabilities Theory (resilience, crisis, pandemic)	67	0.26	0.8382	0.1486
54	Transaction Cost Economics	65	0.25	0.9259	0.1456
55	Agency Theory (ownership, investors, governance, performance)	63	0.24	0.8273	−0.1454
58	Upper Echelons Theory (diversity, TMT, gender/cultural/age diversity)	61	0.24	0.8421	0.2838
61	Stakeholder Theory	59	0.23	0.8439	0.0003
67	Institutional Theory (political connections, lobbying, government)	56	0.22	0.8584	0.089
70	Natural Resource-Based View	55	0.21	0.8061	−0.1319
74	Institutional Theory (Islamic banks, shariah, compliant)	49	0.19	0.8514	−0.0891
75	Resource-Based View + Real Options Theory (venture, VC, CVC, investors)	49	0.19	0.8953	0.1155
109	Ambidexterity Theory	33	0.13	0.9471	0.5814
117	Agency Theory (franchising + knowledge transfer)	31	0.12	0.9265	0.4429
124	Upper Echelons Theory (CEO succession, turnover, founder, interim)	27	0.1	0.818	−0.1646
127	Behavioral Theory of the Firm (slack, organizational)	26	0.1	0.863	−0.1323
132	Agency Theory (audit committee, internal audit, ac effectiveness, audit quality)	25	0.1	0.8323	−0.2171
136	Dynamic Capabilities Theory (Enterprise Resource Planning, ERP)	24	0.09	0.8138	−0.1031
138	Transaction Cost Economics (TCE) + Resource-Based View (RBV)	24	0.09	0.8623	−0.0441
155	Transaction Cost Economics	20	0.08	0.9138	0.1413
163	Natural Resource-Based View (NRBV)	18	0.07	0.8147	−0.2133
183	Resource-Based View (RBV, Inventory Management, Leanness)	14	0.05	0.9509	0.3107
184	Institutional Theory	14	0.05	0.9493	0.3094
207	Real Options Theory + RBV	12	0.05	0.8156	−0.0507

reports prevalence (Size_docs; Prevalence_pct) and coherence (C_V; C_NPMI). Overall, topics exhibit a high C_V (~0.81–0.96) and C_NPMI ranging from −0.22 to 0.44, providing a quality screen for interpretation and determining node sizes in the network.

4.5. Conclusions

This section systematically identified and validated 33 high-quality topics from 25,907 strategic management documents using the BERTopic model. These topics were selected based on superior semantic coherence scores and further enriched by mapping them into a theoretical hierarchy spanning Root Disciplines, Fields, and Subfields. The analysis confirms that the BERTopic model outperforms traditional methods like LDA in generating interpretable and theory-aligned topics. The resulting framework not only captures the diversity and depth of strategic management research but also provides a robust foundation for analyzing topic evolution across historical eras.

5. Discussion

This section interprets the empirical findings from Section 4 in relation to the theoretical framework outlined in Section 2. It addresses the three research questions by clarifying the significance of the extracted topics, evaluating their alignment with the Root Discipline–Field–Subfield structure, and assessing the value added by BERTopic. The discussion emphasizes how these results contribute to understanding the semantic architecture and theoretical development of strategic management.

5.1. Unveiling the Semantic Landscape of Strategic Management: From Core Theories to Emerging Themes

The 33 topics extracted from over 25,000 documents reveal the semantic depth and diversity of the strategic management field. These topics reflect both established theoretical constructs—such as Agency Theory, Dynamic Capabilities, and Resource-Based View—and emerging domains, such as digital platform ecosystems and stakeholder capitalism.

Rather than simply clustering keywords, the results uncover distinct conceptual boundaries and intersections. For instance, topics on ambidexterity, organizational identity, and ESG discourse suggest the field’s broadening concern with dynamic adaptation and societal relevance. This confirms the conceptual pluralism noted in Section 2, where strategic management was described as evolving through mechanisms of convergence and differentiation.

The high coherence scores validate that the semantic clusters are not arbitrary but represent meaningful intellectual themes. These clusters offer empirical grounding to previously abstract theoretical assumptions and demonstrate how fragmented discourses coalesce into identifiable knowledge structures.

5.2. Theoretical Anchoring Through Multidisciplinary Mapping

Mapping the 33 topics onto the Root Discipline–Field–Subfield framework (as theorized in Section 2) reinforces the epistemological diversity of strategic management. Topics align with economics-based theories (e.g., Agency, TCE), sociology-based frameworks (e.g., Institutionalism, Resource Dependence), psychology-rooted concepts (e.g., Organizational Learning, Upper Echelons), and hybrid management approaches (e.g., Dynamic Capabilities, Ambidexterity).

This alignment suggests that the field does not evolve linearly within disciplines, but rather through overlapping trajectories across disciplines and subfields. For example, the presence of strategy-as-practice and behavioral strategy topics reflects sociology and psychology influences operating alongside economic rationality.

The mapping also clarifies the extent of theoretical integration. Topics such as ecosystem strategy and digital innovation show how contemporary issues draw upon multiple Root Disciplines, reinforcing the Fish-Scale Multiscience model. This correspondence affirms that the extracted semantic structure reflects not just linguistic co-occurrence, but genuine theoretical architecture grounded in epistemic traditions.

5.3. Advancing Semantic Precision and Theoretical Differentiation Through BERTopic

Compared to traditional topic models, the results produced by BERTopic provide greater resolution in detecting subtle theoretical themes and overlaps. Topics are more granular, allowing for the distinction between, for example, different applications of Agency Theory (Board Governance vs. audit structures), or various streams within dynamic capabilities (knowledge orchestration vs. environmental dynamism).

More importantly, BERTopic allows for the emergence of topics not predefined in the search criteria, such as identity strategy, ESG integration, and stakeholder tensions—demonstrating its capacity to surface latent themes that are conceptually grounded yet often overlooked.

This semantic clarity enhances theory mapping. By grounding topic labels in high-weight keywords and aligning them with theoretical categories, the model avoids superficial interpretations. As a result, the findings offer empirical input into theory refinement—clarifying boundaries, suggesting integration opportunities, and identifying areas of fragmentation.

5.4. Contribution to Understanding Strategic Theory Evolution

The results extend the scientific knowledge evolution framework introduced in Section 2 in several ways:

• Structural Validation: The successful mapping of semantic topics to theoretical layers confirms that strategic management is not merely a fragmented field, but one with identifiable, structured, and evolving knowledge domains.
• Mechanisms of Evolution: The observed topic dynamics—such as convergence of digital strategy with organizational learning, or differentiation within governance models—mirror the evolution mechanisms discussed in Section 2, namely theoretical convergence, differentiation, and regeneration.
• Clarifying Theoretical Ambiguity: By exposing overlapping topics and hybrid constructs, the findings offer tools for resolving ambiguities in the field’s theoretical core. For instance, the separation of platform strategy from traditional resource-based themes reflects an emerging shift toward ecosystemic thinking.
• Future Synthesis: The identified topic clusters and their temporal trajectories provide a data-driven basis for future synthesis, allowing scholars to track theoretical lineage, anticipate convergence points, and identify gaps in conceptual coverage.

5.5. Application for Empirical Decision-Making (BSC Translation)

To make the results actionable, we translate selected topics into Balanced Scorecard (BSC) strategy maps. The translation links each topic’s salient concepts (from c-TF-IDF keywords and exemplar documents) to objectives in the Learning a d Growth, Internal Process, Customer, and Financial perspectives while preserving its disciplinary provenance (economics–management–sociology).

Table 7. Mapping of 33 strategic management topics to the strategy map’s four perspectives and Root Disciplines.

Strategy Map	Economics	Management	Sociology
Financial Perspective	Behavioral Agency Theory Agency Theory (Board Governance) Agency Theory (Executive Compensation) Agency Theory (ownership, investors, governance) Agency Theory (franchising and knowledge transfer) Agency Theory (Audit Committee Effectiveness) Agency Theory (corporate governance, board, ownership) TCE (Cross-Border M&A Strategy) TCE (M&A Performance and Valuation)
Customer Perspective		Stakeholder Theory	Institutional Theory (Political Ties Strategy) Institutional Theory (Islamic Finance) Legitimacy Theory Institutional Theory (ISO Legitimacy Strategy)
Internal Process Perspective	Knowledge-Based View (KBV) Resource-Based View (Intellectual Capital Efficiency) Resource-Based View + Real Options Theory (venture, VC) TCE + RBV (M&A Strategy) Resource-Based View (Inventory Leanness Strategy) Real Options Theory + RBV (Investment Strategy) Natural Resource-Based View (Green Supply Chain) Natural Resource-Based View (Sustainable Supply Chain)
Learning and Growth Perspective		Dynamic Capabilities Theory (Crisis Resilience) Dynamic Capabilities Theory (ERP Implementation)

provides the crosswalk between topics and BSC perspectives; below, we offer a hypothetical strategy map derived from the findings, addressing the reviewer’s request. The mapping follows the reproducible procedure detailed in Section 3.4 and Section 3.5 (signal extraction → BSC codebook alignment → dual-rater adjudication).

Illustrative strategy map: Agency Theory–Board Governance.

High-weight terms (board independence, audit committee, monitoring, incentives, ownership) indicate an economics-rooted focus on mitigating agency costs, complemented by management routines and sociological legitimacy concerns.

• Learning and Growth: Build board governance capability.

Objectives: director training on fiduciary duties; analytics for board materials.

KPIs: % directors completing annual training; timeliness/coverage of board dashboards.

2.

Internal Process: strengthen monitoring and control routines.

Objectives: audit committee cadence and scope; whistle-blowing and risk escalation workflow.

KPIs: audit committee meetings/year; remediation cycle time for material weaknesses; % agenda items closed.

3.

Customer (capital market): increase investor confidence and governance signals.

Objectives: governance pillar score; analyst sentiment; litigation/violation incidence.

KPIs: MSCI/Refinitiv “G” score; abnormal bid–ask spread around earnings; governance-related lawsuits.

4.

Financial: reduce agency costs and stabilize earnings.

Objectives: lower cost of equity; improve earnings quality and valuation multiples.

KPIs: implied cost of equity; discretionary accruals; Tobin’s Q/EV-to-EBIT.

The causal chain Learning and Growth → Internal Process → Customer → Financial articulates how governance capabilities diffuse into routines, generate credible market signals, and ultimately affect financial outcomes. The same procedure can be applied to other topics (e.g., Socioemotional Wealth → Family Governance Capability; RBV–Intellectual Capital → Knowledge Orchestration Routines).

6. Conclusions and Future Directions

How we could systemize fragmented knowledge seems to be the most important question in the knowledge-bursting world. Strategic management’s knowledge base is fragmented, and bag-of-words maps lose the latent information we could obtain. To address this, the study adopts a semantics-first design and analyzes 25,907 WoS records from 1980 to 2025 after screening out items lacking titles, keywords, or abstracts. BERTopic (Sentence-BERT all-MiniLM-L6-v2 → UMAP → HDBSCAN) discovered 33 coherent topics, labeled them with a rubric, and aligned them to a Fish-Scale hierarchy (Root → Field → Subfield). The longitudinal map shows a migration from industry analysis toward RBV/KBV, dynamic capabilities, and digital ecosystem logics. Bridging themes connect managerial and technological work. Compared with frequency models, semantic maps draw crisper boundaries, keep labels stable, and surface cross-topic recombination that informs theory and practice.

Quantitatively, coherence improves with embeddings—the overall BERTopic C_V = 0.6620 versus LDA C_V = 0.4359—and every retained topic exceeds ≈0.80. The Fish-Scale mapping clarifies disciplinary provenance and overlap, while a Balanced Scorecard translation demonstrates how topics become objectives, processes, customer outcomes, and financial KPIs. Empirically, the BSC mapping produces KPI scaffolds usable for portfolio review and capability audits. Limitations remain. Metadata-based screening can under-label rare micro-topics; the results are sensitive to encoder and clustering choices; English-dominant indexing may bias coverage. Coherence and coverage audits mitigate, but expert judgment is still required for labels and boundaries.

The next steps extend both scholarship and application. On the academic side, we must tighten temporal resolution to quantify emergence, differentiation, convergence, and extinction; link semantic proximity to citation trajectories through main-path analysis on embedding graphs; and formalize provenance-tracked, LLM-assisted labeling with saliency, and robustness audits. Cross-database and multilingual tests (Scopus, Dimensions; domain-specific encoders) will probe generalizability. On the empirical side, we must connect topic intensities and centrality to firm outcomes and policy indicators via panel models; run pilots where topic-derived capability maps drive Balanced Scorecard objectives, process controls, and leading indicators; and evaluate lift through pre/post or A/B designs. To support adoption, we must deliver interactive dashboards for monitoring and alerts, API hooks for refresh, and reproducible configs. External validation across sectors will test transportability and calibrate decision thresholds. Horizon scanning can flag inflection points when topic velocities or recombination rates cross bands. Together, these efforts build a cumulative research program while turning the map into a planning instrument.