Sustainability-Oriented Assessment of the Governance Capacity of China’s Intangible Cultural Heritage Policies: A Hybrid BERTopic-PMC Systematic Analysis

Abstract

Achieving the Sustainable Development Goals requires intangible cultural heritage (ICH) policies that safeguard living heritage while enabling inclusive participation, capacity building, responsible use, and resilience. We introduce a reproducible, text-as-data framework that links semantic-agenda discovery to mechanism-level policy diagnosis by coupling BERTopic with the Policy Modeling Consistency (PMC) index. Using a corpus of 26 national-level Chinese ICH policies, BERTopic reconstructs agenda structure and stage-wise shifts; PMC operationalizes sustainable governance through nine primary dimensions and 45 binary indicators to test whether policies encode long-horizon, monitorable, and feedback-enabled governance loops. The agenda is dominated by inventory-based listing and key-item safeguarding, then expands toward standardization, implementation coordination, and communication/participation, and recently pivots toward resource allocation and performance governance. Policy design quality is moderate (mean PMC = 6.15; 8 Good, 14 Average, 4 Poor). Strengths concentrate on stated goals, instrument mixes, and implementation arrangements, whereas recurrent gaps cluster in time institutionalization (timelines, milestones, rolling revision), legal-institutional enforceability, and operational monitoring-evaluation-feedback provisions. We translate these bottlenecks into an SDG-aligned upgrading pathway that hardens temporal targets, binds evaluation to budgets and accountability, and codifies participation rights and risk boundaries, enabling ICH safeguarding to shift from project-style action to sustainable, cross-cycle governance.

1. Introduction

As global commitments to cultural diversity and sustainable development intensify, safeguarding intangible cultural heritage (ICH) is no longer a narrowly defined cultural-sector concern. Instead, it is increasingly embedded in the holistic development agenda advanced by the United Nations Sustainable Development Goals (SDGs), emerging as a policy domain that links inclusive social development, place-based economic trajectories, and governance resilience [1,2,3,4,5]. Accordingly, the policy rationale for ICH safeguarding has expanded from protecting heritage “as such” to building governance capacity that can sustain transmission over time. This shift foregrounds long-term institutional arrangements, cross-sector coordination and multi-actor co-governance, stable resource guarantees, and closed-loop systems for monitoring, evaluation, and feedback-driven policy revision. Collectively, these mechanisms enhance policy continuity and adaptability and, in turn, support sustained transmission and social value realization under the pressures of modernization [6,7,8]. Yet modernization simultaneously compresses the socio-ecological conditions of transmission and amplifies pressures for development-oriented use [9], intensifying the inherent tension between safeguarding and utilization, increasing the complexity of resource allocation and actor coordination [10], and, critically, undermining the formation of effective monitoring-and-evaluation feedback loops [11,12]. Therefore, whether the policy system can institutionalize durable, implementable, monitorable, and feedback-enabled arrangements—thereby converting “sustainable development” from a normative aspiration into a testable, correctable governance practice—constitutes a decisive prerequisite for sustainable ICH transmission and for meaningful alignment with the SDG agenda.

As of December 2025, China has 45 elements inscribed on UNESCO’s ICH lists, ranking first worldwide [13]. Domestically, the national inventory of representative ICH items includes 1557 entries and 3610 sub-items, spanning ten categories such as oral traditions, traditional music, traditional theater, and folk customs [14]. Facing a vast, heterogeneous, and regionally uneven ICH landscape, China has progressively consolidated a relatively systematic policy architecture: it is anchored in the Law of the People’s Republic of China on ICH, steered by medium- to long-term plans and targeted policy programs, and operationalized through an array of administrative measures and implementing notices. This architecture provides an institutional foundation for routinized, institutionalized safeguarding [15]. Crucially, under the SDG discourse that elevates cultural sustainability and inclusive development, ICH governance has become increasingly interwoven with rural revitalization, employment and income generation, public cultural-service provision, and digital documentation and dissemination [16]. As a result, the policy system is now expected not only to protect and transmit heritage, but also to manage sustainability-relevant trade-offs, including developmental coordination, responsibility boundaries, and risk governance, arising from expanding modes of use.

However, despite a growing body of scholarship on safeguarding and transmission mechanisms [17,18], the tension between protection and utilization [19], local governance practices [20,21], culture-tourism integration [22,23], and digital communication [24,25], systematic evaluations of policy texts remain strikingly underdeveloped. More specifically, the literature still lacks an assessment framework oriented toward sustainability governance capacity that enables structured comparison and reproducible diagnosis across policies. This gap is not merely methodological; it weakens holistic identification of institutional strengths and bottlenecks, and it constrains evidence-based policy refinement and capacity building aimed at enhancing governance resilience and aligning national ICH governance with the SDG agenda.

To close these gaps and respond to the SDG-era demand for policy-relevant evidence on sustainable heritage governance, this study integrates BERTopic topic modeling with the Policy Modeling Consistency (PMC) index. We first systematically map the agenda structure of China’s national-level ICH policies and trace its stage-wise evolution; we then conduct a multidimensional, quantitative assessment of the policy system’s sustainable-governance capacity, enabling a diagnosis of the structural strengths and bottlenecks that condition cultural sustainability. On this basis, the paper advances policy optimization recommendations and governance capacity enhancement pathways aligned with cultural sustainability goals, and provides a reusable analytical framework and empirical evidence to strengthen long-term institutionalization, monitorability, and adaptability in the national ICH policy system.

To complement the overall research motivation with testable expectations, and based on the literature on ICH governance and the staged evolution of national policy instruments, we propose the following hypotheses to be examined in the empirical analysis (the specific research questions are formulated in Section 2.4). H1: National ICH policies issued in the 2021–2025 stage will exhibit higher overall PMC scores than those in earlier stages, reflecting strengthened system-level governance. H2: Post-2021 policies will show higher scores in sustainable-governance assessment (X8) than pre-2021 policies, indicating enhanced emphasis on sustainability-oriented mechanisms (e.g., participation, risk-bounded utilization, and adaptive monitoring). H3: Despite overall improvements, policy timeliness (X2) will remain a comparatively weaker dimension across stages due to uneven codification of milestones, validity arrangements, and rolling-revision mechanisms in national-level texts.

2. Literature Review

2.1. Sustainable Governance and the Policy Turn in ICH Governance

Sustainable governance is widely framed as a contemporary governance paradigm [26] centered on balancing economic development, social equity, and environmental protection [27]. Under conditions of uncertainty and multiple objectives, it prioritizes long-horizon orientation and intergenerational responsibility [28,29] alongside cross-sector coordination and multi-actor participation [30,31]. Crucially, it requires institutionalizing a closed governance loop through explicit responsibility allocation, deliberate resource provisioning, and monitoring-evaluation-feedback adjustment mechanisms [32]. Building on this foundation, collaborative governance scholarship argues that effective governance depends on governments and societal actors co-deliberating within institutionalized arenas, co-producing arrangements, and sharing responsibilities [33,34], with learning and adaptive adjustment serving as core mechanisms of sustained improvement [35]. These ideas have been incorporated into domains such as environmental/natural resource governance and urban governance [36,37,38,39], where research has shifted from single-sector regulation toward systemic pathways emphasizing policy mixes, coordinated action, and performance feedback [40,41,42]. This shift underscores that sustainable governance must be embedded in executable policy instruments and evaluable institutional arrangements rather than remaining at the level of normative declarations. Recent heritage and tourism governance scholarship further suggests a “regenerative shift,” in which policy goals move beyond sustaining heritage conditions toward active renewal and restoration of cultural assets and the socio-cultural systems that carry them, whether tangible or intangible [43]. In this view, policy effectiveness depends not only on preservation-oriented safeguards but also on repair- and revitalization-oriented instruments that can reproduce cultural value across cycles and territories.

Applying sustainable governance to ICH requires engaging the object’s distinctive attributes. ICH is a “living heritage” maintained through the continuous practice of communities and bearers, combining public significance with place-based expression [44,45]. This “living heritage” attribute also provides a theoretical explanation for why feedback loop failure is more likely in ICH governance than in domains such as environmental policy. Because ICH is sustained through ongoing practice by bearers and communities, many governance outcomes are processual, tacit, and context-dependent (e.g., skills, embodied knowledge, ritual competence, community recognition) [46], which makes them difficult to translate into stable and standardized monitoring indicators. In addition, policy effects often unfold with long time lags and complex attribution pathways, since changes in practice may be shaped simultaneously by market pressures, demographic mobility, tourism, education, and digital mediation. As a result, monitoring signals can become delayed or ambiguous, weakening evaluation feedback and making adaptive correction harder unless policies institutionalize practice-sensitive and participatory monitoring arrangements.

Recent scholarship highlights a “regenerative shift” in heritage and tourism policy, moving beyond sustainability to emphasize active renewal and restoration alongside safeguarding, including for intangible heritage [47]. This view reinforces that ICH policy should be assessed not only for protection intent but also for its capacity to enable ongoing revitalization. UNESCO’s 2003 Convention stresses respect for relevant communities, groups, and individuals, and calls for strengthening safeguarding capacity and public awareness through institutionalized mechanisms [48], providing an international normative basis for policy-oriented ICH governance. However, research warns that administrative rationalization and labeling, although potentially improving coherence, may displace meanings and generate participation imbalances [49,50]. In China, scholarship has mainly traced governance evolution through institutional/spatial approaches [51,52], discourse analysis [53,54,55], and actor-network perspectives [56]. Despite these advances, two gaps remain: first, the literature is dominated by qualitative and case-based work, with limited reproducible and comparative assessment of policy system structure and design quality; second, sustainable governance is often invoked rhetorically rather than operationalized into observable policy components within a systematic evaluative framework. As a result, it remains difficult to assess whether ICH policies institutionalize the durable, collaborative, and resource-backed closed loops needed for long-term governance, including monitoring, evaluation, and feedback-driven adjustment.

2.2. China’s ICH Policy System

China’s ICH governance has been institutionalized through a gradual trajectory, moving from the uptake of international norms to domestic institutional construction and, subsequently, to a more consolidated policy architecture. It has progressively crystallized into a national system centered on the Law of the People’s Republic of China on ICH, structured around ICH resource surveys, representative-item inventory management, transmission and dissemination system building, and the specification of legal responsibilities [57,58]. In parallel, medium- to long-term plans, exemplified by the 14th Five-Year Plan for ICH safeguarding, together with sectoral policies, have broadened governance objectives beyond safeguarding per se to include systematic safeguarding, capacity building, public cultural-service provision, and the mobilization of societal participation [59]. Collectively, these instruments provide an institutional foundation for systematic and comparative analysis of national ICH policies, including regulated dissemination and utilization [60].

Building on this architecture, scholarship has examined governance evolution along several strands. First, studies of institutional change argue that, following the Convention framework, ICH governance has become increasingly formalized and administratized through inventory-making and listing processes [61,62,63]. Second, a large literature interrogates tensions between safeguarding and utilization, particularly in relation to tourism development [64], commercialization, and place-based lived experience [65,66,67]. Beyond the safeguarding-utilization debate, recent tourism sustainability research highlights the role of corporate social responsibility (CSR) and its links to perceived service quality and resident/tourist satisfaction. Empirical evidence suggests that greater awareness of sustainability and CSR initiatives is associated with tourist preferences and satisfaction, while differences between expected and perceived service quality and demographic factors shape how such initiatives are evaluated [68]. This perspective complements ICH policy analysis by underscoring that implementation outcomes may depend not only on policy design but also on service-delivery practices in tourism-related contexts. Converging on the concern that market logics may intensify spectacularization and homogenization risks. Third, research on community participation and empowerment has grown, documenting how local communities’ roles in listing applications [69], project operation [70], and benefit distribution are often constrained by resource scarcity, limited institutional channels, and asymmetries in discursive power, especially in peripheral regions [71]. Finally, digital transformation has shifted attention to digital documentation, platform-based dissemination, and generative tools, highlighting their implications for heritage representation and governance boundaries and raising new issues of authenticity protection, visibility politics, and inclusive data governance [72,73,74]. Despite these insights, systematic evaluation of policy texts remains underdeveloped: the literature still relies heavily on normative interpretation, policy histories [75,76], or case-based analyzes, and lacks reproducible [77], corpus-based comparison grounded in large-scale textual evidence.

2.3. Public Policy Text Mining and Evaluation

With the rapid expansion of policy document corpora, public policy research has increasingly developed quantitative approaches grounded in the premise that “text is data” [78]. These methods aim to detect shifts in policy agendas, instruments, and governance priorities while reducing the selective bias and limited reproducibility associated with purely manual interpretation [79]. Common approaches include dictionary-based techniques and content analysis [80], supervised classification [81], and unsupervised learning methods typified by topic modeling [82]. Topic models are widely used because they can extract latent agenda structures without imposing an a priori framework [83]. More recently, semantic embeddings and pretrained language models have moved topic modeling beyond co-occurrence statistics toward semantics-driven clustering [84]. BERTopic, which embeds documents, clusters them, and derives topic representations using class-based TF-IDF, can generate coherent topics without pre-specifying topic numbers and is robust to noise and text-length variation. These properties make it well suited to policy corpora characterized by cross-sector terminology and compound issue formulations [85]. However, topic models remain primarily descriptive; they indicate what policies discuss and how balanced agenda coverage is, but they do not by themselves diagnose whether policy design is sufficiently complete or whether implementation is supported by a closed-loop governance architecture.

In policy evaluation, assessments are commonly classified along the policy cycle into ex ante evaluation before implementation, process evaluation during implementation, and ex post evaluation after implementation [86]. Ex ante evaluation is widely argued to intervene early in program design to strengthen intervention logic and improve policy quality [87]. In ICH-related research, evaluations more often concentrate on ex post impacts and performance, for example, through indicator systems that measure the economic, social, and environmental effects of heritage projects or policy implementation [88]. Although outcome-focused approaches capture realized effects, they typically rely on existing datasets and observed results, and therefore provide limited capacity to diagnose the completeness of institutional design before a policy is issued. Against this background, the PMC index model proposed by Ruiz Estrada has increasingly been used for text-based ex ante evaluation in public policy studies [89]. Using policy documents as the primary evidence base, the PMC model structures and quantifies key elements, including policy objectives, instrument configurations, resource guarantees, implementation arrangements, and monitoring-and-evaluation provisions. This enables assessment of design consistency and balance even when comparable performance indicators are unavailable, and it visualizes strengths and weaknesses in an interpretable form. Owing to its generalizable yet customizable architecture, the PMC model has been applied across domains such as industrial policy, budget performance management, land-use governance, and ecological and environmental policy.

However, the PMC approach remains sensitive to researcher judgment during indicator construction and binary coding, and it can be biased toward high-frequency textual cues, creating risks of omitted variables and the compression of meaningful nuances [90]. To address these limitations, recent studies have incorporated text mining and topic modeling to provide agenda cues that complement the PMC framework and to embed specific evaluative orientations that link agenda identification with structural diagnosis [91]. Nevertheless, the literature still lacks a systematic integration of semantically coherent topic identification methods with the PMC framework capable of supporting reproducible, comparable ex ante evaluation of national-level ICH policies and delivering sustainability-oriented governance diagnostics.

2.4. Research Questions

Building on this analytical approach, this study addresses three questions:

RQ1: What agenda structure and priority issues characterize China’s national-level ICH policies, and how do agenda configurations vary across policy stages?

RQ2: From a sustainable-governance perspective, how consistent and balanced is the design of national-level ICH policies overall, and how do key dimensions perform?

RQ3: Combining agenda structure mapping with PMC-based evaluation, where are the main structural bottlenecks in sustainable-governance capacity, and how should policy instrument mixes and institutional arrangements be optimized?

3. Research Design

3.1. Data Sources and Preprocessing

Systematic collection and standardized processing of policy documents underpinned the text-mining and quantitative evaluation procedures [92]. This study examined China’s national-level ICH policies. Policy texts were retrieved from authoritative sources, including the Chinese Government website, the Ministry of Culture and Tourism website, the National Cultural Heritage Administration website, and the PKULaw database. Using keywords such as “intangible cultural heritage”, “ICH safeguarding”, and “ICH inheritors”, we conducted on-site and full-text searches, followed by cross-source verification and deduplication.

To ensure relevance and comparability, we applied four screening criteria. (1) Issuing authority: Only centrally issued documents (e.g., the National People’s Congress and its Standing Committee, the State Council, and State Council ministries/commissions) were retained. (2) Document type: We prioritized binding legal/regulatory and normative policy instruments with explicit implementation rules (e.g., laws/administrative regulations, departmental rules, normative documents), and excluded primarily informational texts (e.g., approval replies, inventory announcements, selection/commendation notices). (3) Substantive scope: Documents had to directly address ICH safeguarding, transmission, regulated utilization, or governance mechanisms; broad policies with only general ICH mentions and no ICH-specific arrangements were excluded. (4) Validity/version control: Only currently in-effect texts were kept, and only the latest valid version was retained when multiple versions existed. This workflow yielded 101 initial documents and a final analytical sample of 26, which were standardized into a unified format and annotated with metadata (Supplementary Materials Section S1).

For topic modeling, we converted the 26 documents to plain text and segmented them into paragraphs to reduce long-document effects, generating 2386 paragraph units. We then cleaned and normalized the corpus by removing punctuation and numeric noise, performing Chinese word segmentation with a domain-specific custom dictionary, merging synonyms to harmonize terminology, and removing low-information tokens using public and self-compiled stop-word lists. During this step, many paragraph units were excluded because they mainly contained standard administrative/legal boilerplate (e.g., issuing formulae, addressee and copying/archival notes, signatures, lists of Attachments), repeated citation blocks duplicated across documents, or extremely short fragments that did not meet the minimum token threshold after cleaning. Finally, we filtered tokens using a term-frequency proportion threshold of ≥2‰ (occurrence ≥ 4), retaining 718 valid terms, and constructed 758 paragraph-level samples for subsequent BERTopic modeling (Figure 1). Importantly, this filtering was applied only to strengthen the semantic signal for topic discovery; the subsequent policy evaluation was conducted using the full text of all 26 policy documents. Therefore, document-level typological information (e.g., document type, issuing authority, and formal metadata) and the corresponding evaluation judgments based on such information were not affected by the paragraph-level filtering.

3.2. Evaluation Framework Design

This study assessed the sustainable-governance capacity of China’s national-level ICH policies and developed an integrated workflow that moved from macro-level agenda mapping to micro-level diagnostic evaluation (Figure 2). At the macro level, we performed policy text mining using national ICH policy documents as inputs. We cleaned the texts, conducted word segmentation, extracted high-frequency terms, and applied co-occurrence network analysis to identify core concepts and their associative structures. In parallel, we applied BERTopic to the corpus to extract topics and produce a keyword-represented agenda taxonomy, which was used to characterize policy attention structure and theme distribution. At the micro level, we constructed a PMC indicator system and defined scoring rules for secondary variables to enable structured quantitative evaluation. We then calculated PMC indices and visualized PMC surfaces to pinpoint strengths and weaknesses in policy design.

3.3. Research Methods

3.3.1. BERTopic Model

We applied BERTopic, a BERT-based topic modeling framework that combined pretrained language models with document-embedding generation, dimensionality reduction, and clustering to identify latent semantic topics in the policy corpus. The model also enabled temporal analysis by tracking changes in topic distributions over time, which allowed visualization of time-varying topic prevalence and quantification of fluctuations in policy attention. Compared with conventional Latent Dirichlet Allocation, BERTopic used Transformer-based contextual representations to better capture semantics and reduce polysemy-driven ambiguity, thereby improving topic identification. Its pretrained parameters supported scalable processing of large corpora, and its density-based clustering inferred topic numbers with limited manual intervention, improving efficiency for large-scale policy text analysis (Figure 3). For transparency and reproducibility, we provide the embedding model specification, the UMAP/HDBSCAN hyperparameter settings, and the grid-search selection procedure in Supplementary Materials Section S2.

3.3.2. Construction of the PMC Index Model

We employed the PMC Index model to quantitatively evaluate ICH policy texts. The PMC index is a text-based policy evaluation approach widely used to assess design consistency and effectiveness. Proposed by Ruiz Estrada [89], it was grounded in the Omnia Mobilis assumption, which treated policy elements as interdependent and dynamic and therefore evaluated them jointly rather than in isolation. Accordingly, the PMC model used a two-tier variable structure, typically including up to ten primary (first-level) variables and an unrestricted set of secondary variables, with equal weights assigned across variables. This structure was intended to capture heterogeneity in policy design and support comparative diagnosis of strengths and weaknesses. The PMC construction framework applied in this study is presented in Figure 4, and the procedure followed the steps below.

1. Text Selection and Preprocessing

To satisfy the PMC model’s requirement for assessing the completeness of policy elements, we evaluated all 26 screened national-level ICH policy documents, covering laws/regulations, plans, and normative policy instruments. We adopted a full-corpus design rather than sampling to reduce selection bias and strengthen the robustness of cross-policy comparisons, thereby supporting subsequent quantitative evaluation and mechanism-oriented interpretation.

2. Variable Specification and Parameter Identification

Based on the BERTopic outputs and prior scholarship, we operationalized a set of quantitative indicators for policy evaluation. Following the variable-construction logic of Ruiz Estrada [93] and Zhang Yong’an [94], we organized indicators using a policy-typology scheme and built a sustainability governance-oriented framework with nine primary variables and 45 secondary variables. The primary variables comprised policy legal status (X1), policy time horizon and validity (X2), policy target groups (X3), policy perspective (X4), policy focus (X5), policy instruments (X6), content quality assessment (X7), sustainable-governance assessment (X8), and policy safeguards (X9). Variable definitions and parameter identification rules are summarized in

Table 1. Evaluation indicators at all levels, identification rules and source basis.

Primary Variable	Secondary Variable	Definition Observation Point	Coding Criterion
X1 Policy legal status	X1-1 Law	Laws, administrative regulations, ordinances	The title or document number clearly indicates “Law/Ordinance/Administrative Regulation”, etc.
	X1-2 Ministerial regulation normative document	Ministerial rules, administrative measures, provisions	The text is titled “Measures/Provisions/Rules/Implementing Rules”, and carries general binding force.
	X1-3 Plan outline scheme	Medium- to long-term plans; action schemes	The text is titled “Plan/Outline/Scheme/Action Plan” and specifies a task framework.
	X1-4 Opinions notices guiding document	Policy orientation and task deployment	The text is titled “Opinions/Notice/Guiding Opinions/Measures”, etc.
	X1-5 Policy interpretation guideline	Interpretive guidance on implementation	The text is titled “Interpretation/Q&A/Guideline/Guide” and corresponds to a specific policy document.
X2 Policy time horizon and validity	X2-1 Explicit start-end dates or validity period	Validity; implementation term (for time-bounded instruments)	For plans/outlines/programs/implementation schemes, the text specifies an explicit validity period or implementation term (e.g., “effective from…until…/valid for … years”). For laws/regulations/normative opinions intended to remain effective until superseded, the absence of an end date is not regarded as a deficiency under this item.
	X2-2 Explicit staged arrangement timetable	Phased goals; milestones (for plan-type documents)	For plans/outlines/programs/implementation schemes, the text specifies time nodes (e.g., “by 2025/2030”), phased tasks, milestones, or annual work plans. For laws/regulations without staged timetables by design, this item is not required.
	X2-3 Explicit dynamic revision rolling update	Post-evaluation revision mechanism (applicable to all document types)	The text specifies periodic review, revision, rolling adjustment, or timely updates (e.g., review cycles, revision triggers, post-evaluation update mechanisms).
X3 Policy target groups	X3-1 Government agencies with explicit responsibilities	Division of labor; responsibility chain	Specific agencies are named and assigned tasks (e.g., culture and tourism, education, finance).
	X3-2 Bearers safeguarding units	Support for key actors	The text specifies bearers, safeguarding units, training bases, etc., with concrete measures.
	X3-3 Communities and civil society organizations	Community and social forces	The text specifies mechanisms/roles for communities, associations, NGOs, volunteer groups, etc.
	X3-4 Market actor culture-tourism enterprises	Industry and utilization actors	Enterprises/platforms/culture-tourism operators are specified with boundaries or responsibilities.
	X3-5 Public school media	Targets of education and communication	The text specifies public education, school curricula, media dissemination arrangements, etc.
X4 Policy perspective	X4-1 Supply-side capacity building	Resources, facilities, talent, platforms	The text provides for funding, institutions, talent, platforms, inventory system building, etc.
	X4-2 Demand-side public services and participation	Public services, cultural consumption, participation	The text specifies public cultural services, education/communication, participation mechanisms, supply-demand matching.
	X4-3 Enabling environment: Institutions and regulation	Standards, oversight, risk governance	The text specifies standards, regulation/enforcement, risk prevention, accountability mechanisms, etc.
X5 Policy focus	X5-1 Survey and inventory management	Census, documentation, dynamic inventory management	The text specifies surveys, databases, nomination/review, exit/adjustment mechanisms, etc.
	X5-2 Transmission system and capacity building	Bearer development; transmission practices	The text specifies bearer accreditation, training, transmission bases, apprenticeship systems, etc.
	X5-3 Safeguarding and responsible utilization	Use boundaries; transformation pathways	The text specifies culture-tourism integration, productization with boundary control, intellectual property, etc.
	X5-4 Communication, education, and public awareness	Education, display, dissemination	The text specifies curricula, performances/exhibitions, media communication, popularization activities, etc.
	X5-5 Digitization and information-based safeguarding	Digital recording; platform governance	The text specifies digital archives, platforms, data standards, digital exhibitions, etc.
X6 Policy instruments	X6-1 Fiscal funding instruments	Special funds, subsidies, service procurement	The text specifies concrete mechanisms such as subsidies, earmarked funds, awards/subsidies, government procurement of services.
	X6-2 Tax and financial support	Tax incentives, credit, funds	The text specifies tax relief, financing/loans, funds, interest subsidies, etc.
	X6-3 Talent and bearer support	Allowances, appraisal, training system	The text specifies bearer subsidies, professional-title appraisal, training programs, etc.
	X6-4 Infrastructure and platform building	Venues, bases, digital platforms	The text specifies building venues/stations, transmission workshops, exhibition centers, digital platforms, etc.
	X6-5 Standards, norms, and guidelines	Standards, specifications, operating procedures	The text specifies standards, guidelines, norms, checklists, etc.
	X6-6 Combined incentives and constraints	Rewards/penalties, accountability, licensing boundaries	The text specifies incentives/penalties, negative lists, sanctions, or accountability.
	X6-7 Information disclosure and public communication	Disclosure; dissemination strategy	The text specifies disclosure arrangements, publicity plans, education systems, etc.
	X6-8 Coordination instruments	Cross-sector linkage; collaboration mechanisms	The text specifies joint conferences, coordination platforms, cross-department task division mechanisms, etc.
X7 Content quality assessment	X7-1 Adequate problem diagnosis	Context, tensions, risks	The text identifies concrete problems (e.g., transmission rupture, commercialization risk, digital risk).
	X7-2 Clear and verifiable objectives	Clear goals/indicators/scope	The text specifies checkable objectives, quantitative indicators, or clearly bounded scope.
	X7-3 Clear implementation pathway	Tasks, steps, responsibility chain	The text specifies task lists, roadmaps, and responsible actors.
	X7-4 Monitoring and evaluation arrangements	Evaluation, supervision, feedback	The text specifies indicators, evaluation frequency, supervision mechanisms, and corrective-action loops.
X8 Sustainable-governance assessment	X8-1 Intergenerational transmission mechanisms	Long-term learning and transmission chain	The text specifies bearer pipeline, apprenticeship, curriculum systems, school-based transmission, etc.
	X8-2 Community participation and co-governance	Multi-actor participation; deliberative mechanisms	The text specifies community deliberation, participatory safeguarding, co-management, benefit negotiation, etc.
	X8-3 Protection of authenticity and integrity	Cultural context and process integrity	The text specifies preventing distortion, maintaining context, safeguarding craft/process integrity, etc.
	X8-4 Sustainable use and risk control	Boundary governance between safeguarding and development	The text specifies tourism carrying capacity, commercialization boundaries, overdevelopment risks, negative lists, etc.
	X8-5 Equity and inclusion	Cultural rights of vulnerable/marginalized groups	The text specifies benefit mechanisms/accessibility for ethnic minorities, rural areas, vulnerable groups.
	X8-6 Digital safeguarding and open sharing	Digital archives and data governance	The text specifies digital archives, open platforms, data standards, copyright and sharing rules.
	X8-7 Monitoring, evaluation, and adaptive governance	Dynamic adjustment and learning mechanisms	The text specifies dynamic monitoring, annual evaluations, rolling revisions, corrective mechanisms, etc.
X9 Policy safeguards	X9-1 Leadership and coordination mechanisms	Lead agency, coordination, linkage	The text specifies leading groups, joint meetings, lead units, coordination mechanisms.
	X9-2 Resource guarantees	Funding, land/space, facilities	The text specifies earmarked funding, resource pooling, space/facility, etc.
	X9-3 Rule of law and accountability	Responsibility, sanctions, accountability	The text specifies legal liabilities, penalties, accountability, audit/oversight, etc.
	X9-4 Data and statistical support	Reporting, information systems	The text specifies reporting systems, statistical indicators, information-system support, etc.
	X9-5 Performance and incentive mechanisms	Performance appraisal; evaluation linkage	The text links appraisal results to resource allocation, project adjustments, incentives, etc.

.

3. Construction of the Multi-Input-Output Table

We constructed a multi-input-output table to enable multidimensional quantitative assessment of each policy by decomposing each primary variable into multiple components. Primary variables were treated as independent and non-sequential, and each was operationalized through equally weighted secondary variables with no fixed upper limit. Based on the predefined variable structure and parameter identification rules, we compiled the multi-input-output table used in this study (

Table 2. Multi-input-output of each data element policy.

Main Indicators	Secondary Indicators
X1	X1:1	X1:2	X1:3	X1:4	X1:5
X2	X2:1	X2:2	X2:3
X3	X3:1	X3:2	X3:3	X3:4	X3:5
X4	X4:1	X4:2	X4:3
X5	X5:1	X5:2	X5:3	X5:4	X5:5
X6	X6:1	X6:2	X6:3	X6:4	X6:5	X6:6	X6:7	X6:8
X7	X7:1	X7:2	X7:3	X7:4
X8	X8:1	X8:2	X8:3	X8:4	X8:5	X8:6	X8:7
X9	X9:1	X9:2	X9:3	X9:4	X9:5

).

4. Computation of the PMC Index

Following Estrada’s procedure, the PMC index was calculated as follows. (1) For each policy text, the value of each secondary variable was determined using binary coding, as shown in Equations (1) and (2), where 1 indicates the presence of the specified criterion and 0 indicates its absence. (2) The score of each primary variable was computed as the arithmetic mean of its constituent secondary-variable scores, as shown in Equation (3), ensuring that each primary-variable score remained within the interval (0, 1). (3) The final PMC index score for each policy was obtained by summing the scores of all primary variables, as shown in Equation (4).

$$X_{ij}\in \left\{0,1\right\}$$

(1)

$$X_i\in \left[0,1\right]\mathrm{i}=1,2,\dots ,9$$

(2)

$$X_i={\displaystyle \frac{1}{n_i}}{\sum }_{j=1}^{n_i}{x}_{ij},i=1,2,\dots ,9$$

(3)

$X_i$ denotes the score of the $i$-th primary variable, $x_{ij}$ represents the $j$-th secondary variable under the $i$-th primary variable (coded as 0/1), and $n_i$ is the number of secondary variables included in the $i$-th primary variable.

$$\mathrm{P}\mathrm{M}\mathrm{C}={\sum }_{i=1}^9{X}_i$$

(4)

Because this study specified nine primary variables, the PMC index score for each policy ranged from 0 to 9. According to Estrada’s grading criteria, scores were classified into four levels (

Table 3. Policy evaluation grades of the PMC index.

PMC Index	0–4.99	5–6.99	7–8.99	9–10
Evaluation Level	Poor	Average	Good	Excellent

). Scores between 8.0 and 9.0 were rated as “excellent”. Scores between 6.0 and 7.9 were rated as “good”. Scores between 4.0 and 5.9 were rated as “acceptable”. Scores between 0 and 3.9 were rated as “poor”, indicating low policy consistency.

To enhance the reliability and reproducibility of PMC index scoring, we implemented a standardized coding protocol to minimize subjective bias. First, we developed a coding manual that operationalized the variables and specified explicit binary decision rules (e.g., “fiscal support” was coded as 1 only when concrete mechanisms such as subsidies or earmarked incentives were stated; principle-only wording was coded as 0). Second, two raters conducted joint pilot coding and revised the manual to harmonize criteria. Third, both raters independently coded all 26 policies (1215 data points) and tested inter-rater agreement using Cohen’s kappa. Disagreements were resolved through consensus discussion and were further verified by an industry practitioner and a policy evaluation scholar, yielding the finalized dataset and PMC results.

5. Visualization Using the PMC Surface

We visualized results using PMC surfaces by transforming each policy’s nine primary-variable scores into a matrix and rendering it as a three-dimensional surface plot. To construct comparable surfaces, we arranged the nine primary variables into a symmetric 3 × 3 PMC matrix as specified in Equation (5), and generated one PMC surface for each policy based on this matrix.

$$PMC\left(Suface\right)=\left[\begin{array}{ccc}X1& X2& X3\\ X4& X5& X6\\ X7& X8& X9\end{array}\right]$$

(5)

3.4. Research Hypotheses

Building on the staged evolution of China’s national ICH governance and the PMC variable system defined above, we propose the following testable hypotheses to reduce the purely exploratory appearance of the study and to clarify whether observed policy quality patterns are expected.

H1 

(overall improvement). National ICH policies issued in the 2021-2025 stage will exhibit higher overall PMC scores than those issued in earlier stages, reflecting the deepening of system-wide governance arrangements and instrument integration.

H2 

(late-stage shift to resource-performance governance). Policies in 2021–2025 will show higher scores in policy instruments and safeguard-oriented governance dimensions (especially the instrument mix and performance/accountability-related provisions), consistent with the empirical trend that policy attention pivots toward fiscal resource allocation and performance governance in the later stage.

H3 

(persistent weakness in time institutionalization). Despite overall improvement, policy time horizon and validity (X2) will remain a comparatively weaker dimension across stages, because many national-level laws/regulations and guiding documents are designed to be effective until superseded and often do not specify explicit end dates or milestone timetables, while rolling revision mechanisms are unevenly codified.

H4 

(still-incomplete closed-loop sustainability mechanisms). Compared with early stages, post-2021 policies will more frequently mention sustainability-relevant mechanisms (e.g., participation, transmission support, risk control, and evaluation), but monitoring-evaluation-feedback and adaptive-adjustment clauses will remain insufficiently institutionalized in a subset of documents, limiting the formation of a robust closed-loop governance architecture.

4. Results

4.1. Macro-Level Analysis of Policy Texts

4.1.1. Term-Frequency Results

Words are fundamental semantic units, and their frequencies indicate policy attention and provide features for subsequent topic modeling and PMC mapping. We screened tokens by document-level prevalence, retaining terms appearing in at least 0.1% of documents (frequency ≥ 4). We then ranked eligible terms by frequency and visualized the top 50 (

Table 4. List of frequent words.

Number	Name	Frequency	Number	Name	Frequency
01	Cultural heritage	851	26	Region	121
02	Material	841	27	Traditional craftsmanship	121
03	Protection	819	28	Administrative department	121
04	Inheritance Project	584	29	Society	115
05	Project	507	30	Disabled person	108
06	Culture	490	31	Tourism Department	104
07	Intangible cultural heritage work	488	32	Establish	102
08	Work	451	33	Funds	96
09	Representativeness	291	34	Utilize	91
10	National level	214	35	Xi Jinping	89
11	Development	212	36	Eecycling Workshop	89
12	Training and Education	207	37	Directory	87
13	Tourism	203	38	Institutions	84
14	Plan	191	39	Student	84
15	Support	186	40	Display	84
16	Construction	178	41	Lineage	83
17	Organization	167	42	Department	82
18	China	140	43	Ministry of Culture	81
19	Management	140	44	Universities	81
20	Traditional	137	45	Formulate	76
21	Country	132	46	System	75
22	Activity	132	47	Publicity	75
23	Unit	131	48	Excellent	73
24	Provincial level	131	49	Skill	70
25	Implementation	131	50	Characteristic	70

; Figure 5). The most frequent terms were “cultural heritage” (851), “material” (841), “protection” (819), “inheritance project” (584), and “project” (507), indicating that policy discourse prioritizes heritage/protection frames and project-based governance. Governance by graded listing was reflected by “representativeness” (291), “national level” (214), “provincial level” (131), and “directory” (87). Capacity-building terms included “training and education” (207), “traditional craftsmanship” (121), and “institutions” (84). Utilization-related terms included “tourism” (203), “utilize” (91), and “display” (84). Multi-actor coordination was indicated by “organization” (167), “administrative department” (121), and “Ministry of Culture” (81). Instrument-oriented terms included “plan” (191), “support” (186), “construction” (178), “implementation” (131), and “funds” (96).

Overall, the high-frequency terms concentrated on “safeguarding” and “transmission”, with governance objects structured through hierarchical listing and classification (e.g., projects, inventories, national-level designations). Capacity-building terms (e.g., training, institutions, trainees, traditional crafts) co-occurred with implementation and resource-related terms (e.g., funding, support, plans, implementation, construction, systems), indicating operational arrangements and resource allocation. Actor- and place-related terms (e.g., administrative departments, Ministry of Culture, organizations, units, regions) reflected multi-level coordination and geographically differentiated implementation. This term-frequency profile informed subsequent BERTopic topic identification and PMC-based quantitative evaluation.

4.1.2. Semantic Network Topic Clustering Analysis

We conducted semantic network analysis based on term co-occurrence to capture relational structure beyond term frequencies. Using the 50 high-frequency terms, we built a co-occurrence network (Figure 6) with dense connectivity. “Cultural heritage”, “material”, “safeguarding”, and “transmission projects” were central nodes and, together with “project” and “work”, formed the narrative backbone. Clustering identified three semantic modules: (1) a safeguarding-utilization module linking “cultural heritage/material/safeguarding” with “representative”, “tourism”, and “utilization”; (2) a governance instruments and implementation module centered on “management”, “planning”, “formulation”, “implementation”, “funding”, and “administrative departments”; and (3) a macro-level governance and system-building module including “system”, “support”, “ICH work”, and “China”, connected to transmission actors and implementation contexts. Core nodes also functioned as cross-module connectors: multiple edges linked “safeguarding”, “cultural heritage”, “management”, and “work”, indicating coupling between goal statements and implementation language. “Tourism/display/utilization” were directly connected to or near “safeguarding”, indicating parallel co-occurrence in policy discourse. This network structure provided inputs for subsequent BERTopic topic identification and PMC dimension mapping.

4.2. Topic Identification and Evolution of China’s ICH Safeguarding Policies

4.2.1. Topic Identification and Validation of Topic Representations

BERTopic allows the number of topics to be generated automatically and adjusted using evaluation metrics [95]. We used a three-step procedure: automatic generation, quantitative comparison, and manual refinement. Candidate models were first generated without fixing the topic number. We then compared coherence across alternative topic counts and considered coherence together with topic coverage and interpretability. A three-topic model achieved the highest coherence (0.598) but was too coarse to represent agenda heterogeneity across governance objects, instruments, and implementation mechanisms. A 21-topic model retained relatively high coherence (0.511) but produced a fragmented and redundant structure. The nine-topic model provided a better trade-off, with coherence 0.482, outperforming adjacent settings (8 topics: 0.467; 10 topics: 0.447) and offering suitable granularity for mapping to PMC dimensions.

The nine-topic solution included an outlier cluster (Topic = −1). which in HDBSCAN denotes stochastic noise or unclustered data points that do not belong to any stable topic. To reduce outlier effects, we compared three topic representation optimization strategies (c-TF-IDF, distributions, embeddings) and selected distributions, which yielded the highest coherence (0.508). After optimization, eight valid topics were retained (

Table 5. Names of Each Topic and Their Data Quantities.

	Topic Name	Data Proportion
Topic 1	International Convention Compliance and Multilateral Governance Mechanisms	123
Topic 2	Inclusive Inheritance and Social Support Systems	33
Topic 3	Financial Resource Allocation and Performance Governance	35
Topic 4	Identification Systems and Standardized Protection Mechanisms	67
Topic 5	Dissemination, Promotion, and Public Participation Mechanisms	77
Topic 6	National List and Key Protection Governance	196
Topic 7	Development Synergy and Poverty Alleviation Support Pathways	33
Topic 8	Integrated Implementation Mechanisms for Protection and Transmission	121

). For each topic, we extracted the top 10 keywords (Figure 7). Keyword validity was checked using topic-wise descending score curves and the mean score curve (Supplementary Materials Section S3): c-TF-IDF scores decreased with keyword rank, and the mean curve showed the same monotonic decline, indicating that higher-ranked keywords provided stronger topic discriminability.

4.2.2. Topic Structure Relations and Semantic Associations

Topic Space Distribution and Hierarchical Clustering Structure

The BERTopic topic-map and hierarchical-clustering outputs jointly described structural relations among the eight topics. In the two-dimensional semantic space, topics were well separated (Figure 8), and bubble sizes matched text volumes, indicating stable agenda stratification. Topic 6 (National List and Key Protection Governance; n = 196) was located near the center, showing that inventory-based governance and key-item safeguarding constituted the main axis of the policy narrative and connected to multiple institutional and instrumental themes. On the right side, Topics 8, 4, and 5 formed a clustered module, indicating frequent co-occurrence among implementation execution, standards and norms, and communication and participation. Hierarchical clustering supported this pattern (Figure 9). Topic 8 and Topic 6 merged first at a low distance, and Topic 4 joined next, indicating strong coupling between inventory governance and implementation mechanisms/standardized identification, and suggesting close interweaving of institutional design and implementation arrangements in the corpus. In contrast, Topic 3 and Topic 1 merged at higher distances, implying greater semantic independence. Topic 2 was positioned near the center and merged into multiple clusters at later stages, consistent with a bridging function across agenda domains. Together, the two visualizations indicate a composite thematic architecture spanning a core governance framework, implementation and regulatory mechanisms, social mobilization and participation, and resource/external mechanism supports, providing a structural basis for mapping topics to PMC sustainability governance dimensions.

Topic Similarity Matrix and Association Strength

We computed pairwise similarities among the eight topics to assess inter-topic linkages (Figure 10). Excluding the diagonal (self-similarity = 1.00), off-diagonal similarities ranged from 0.31 to 0.91, indicating partial overlap in vocabulary and governance logics while preserving topic boundaries. The highest similarity was between International Convention Compliance and Multilateral Governance Mechanisms and Fiscal Resource Allocation and Performance Governance (0.91). This high value reflects the strong coupling in national ICH policies, where normative discourse is frequently operationalized through fiscal and performance instruments; however, the two topics remain functionally distinguishable in their keyword emphases (e.g., Topic 1 features UNESCO, contracting states, committee, conference, and law, whereas Topic 3 centers on funds, budget, project, performance, and Ministry of Finance). Both topics also showed high similarity with National List and Key Protection Governance (0.84) and Inclusive Transmission and Social Support Systems (0.86), indicating that normative institutional discourse aligned closely with performance instruments, inventory governance, and social-support arrangements within the corpus. Integrated Implementation Mechanisms for Safeguarding and Transmission exhibited moderate-to-high similarity with Standardized Identification Mechanisms (0.77), List Governance (0.71), and Performance Governance (0.71), consistent with its linkage across institutional design, implementation, and evaluative support. The lowest similarity was between Communication, Promotion, and Public Participation Mechanisms and Developmental Coordination and Assistance Pathways (0.31), indicating limited overlap between social-mobilization and development-oriented agendas. This similarity ordering provided a quantitative basis for subsequent stage-wise comparisons of topic prevalence and for identifying agenda linkages and succession relations.

Temporal Topic Evolution and Stage-Wise Distributions

We analyzed annual topic proportions and distributions (Figure 11). The series showed a concentration-fluctuation pattern. In 2006, the corpus was highly concentrated, with Topic 1 and Topic 6 dominating. In 2011, Topic 2 had the largest share (0.71; 15 texts). In 2012, 2017, 2022, and 2023, Topic 8 remained high (0.80, 0.80, 0.80, and 0.75), corresponding to four, eight, four, and three texts, respectively. In contrast, 2021 exhibited a multi-topic structure, led by Topic 8 (49 texts; 0.31), Topic 3 (33; 0.21), Topic 6 (26; 0.17), and Topic 7 (21; 0.13). Some years had low document counts, which elevated single-topic shares to 0.75–0.80, indicating concentrated agenda expression. The concurrent prominence of multiple topics in 2021 indicated a more diversified agenda configuration, making it a key reference point for subsequent stage aggregation and topic flow analysis.

To reduce volatility from uneven annual sample sizes and to reflect policy evolution, we grouped 2005–2025 into four stages: 2005–2010 (institutional introduction and system foundation), 2011–2015 (legalization and institutional standardization), 2016–2020 (systematic advancement and strengthened policy instruments), and 2021–2025 (deepened systemic governance and strengthened sustainability orientation). We aggregated topic shares and counts by stage (Figure 12), which showed a clear shift in agenda structure. In 2005–2010, Topic 6 (0.38; 129 texts) and Topic 1 (0.33; 112 texts) dominated. In 2011–2015, attention shifted to Topic 2 (0.32; 15) and Topic 8 (0.26; 12), with Topic 6 (0.17; 8) and Topic 4 (0.11; 5) also present. The 2016–2020 stage exhibited parallel multi-topic growth: Topic 6 remained substantial (0.27; 33), while Topic 4 (0.21; 26), Topic 5 (0.14; 17), and Topic 8 (0.15; 18) increased concurrently. In 2021–2025, growth concentrated in Topic 8 (0.34; 60) and Topic 3 (0.19; 33), with Topic 7 rising to 0.13 (23). Overall, the staged results indicate a transition from early emphasis on inventories, framework building, and international mechanisms toward a more differentiated agenda, followed by increased attention to implementation mechanisms and resource/performance-oriented governance.

4.2.3. Topic Evolution Trajectories and Agenda Succession Paths

We used a Sankey diagram to track topic flows across stages and to identify agenda persistence, divergence, and convergence (Figure 13). From 2005–2010 to 2011–2015, “National List and Key Protection Governance” and “Integrated Implementation Mechanisms” showed the strongest continuity, with self-loops of 0.627 and 0.924, respectively. Both topics also partially shifted toward instrument-oriented themes (e.g., “Performance Governance”), indicating early coupling of institutional frameworks and implementation arrangements with emerging resource- and performance-related considerations. From 2011–2015 to 2016–2020, “International Convention Compliance” and “Inclusive Transmission and Social Support” remained highly stable, with self-loops of 0.933 and 1.000, respectively, suggesting these themes functioned as a persistent policy foundation. In 2021–2025, flows converged toward “Standardized Identification Mechanisms” and “Integrated Implementation Mechanisms” (self-loops 0.729 and 0.697). Flows from “List Governance” to these topics were 0.700 and 0.690, respectively, indicating increased emphasis on standardization and implementation delivery. Meanwhile, “Fiscal Resource Allocation and Performance Governance” received inputs from multiple sources (0.413, 0.389, 0.307), reflecting strengthened integration of resource allocation and performance-oriented governance in the later stage. Notably, these agenda shifts reflect changes in policy attention, whereas the PMC diagnosis evaluates whether such themes are operationalized into implementable and revisable institutional provisions; therefore, the linkage between topic transitions and PMC performance is interpreted as an agenda-mechanism complementarity rather than a one-to-one correspondence.

4.3. PMC-Based Evaluation Results and Analysis

4.3.1. PMC Policy Index

Based on the text-mining outputs and the PMC computation procedure, we constructed a multi-input-output evaluation table and coded all secondary variables for 26 national ICH policies (Supplementary Materials Section S4). Using Equation (4), we calculated a PMC index for each policy and classified them by the predefined grading criteria. Of the 26 policies, 8 were rated Good (P1, P8, P16, P17, P18, P21, P24, P26), 14 were rated Average (P2, P3, P4, P5, P6, P7, P9, P11, P14, P15, P19, P20, P22, P23), and 4 were rated Poor (P10, P12, P13, P25). The mean PMC score was 6.150, corresponding to an Average level, indicating moderate overall policy design quality and element configuration. To examine indicator-level structure, we decomposed performance using secondary-variable scores and visualized underperforming dimensions via PMC surface plots (

Table 6. Calculation results of PMC policy index.

	X1	X2	X3	X4	X5	X6	X7	X8	X9	PMC Score	Rank	Level
P1	0.200	0.667	1.000	1.000	1.000	1.000	1.000	1.000	1.000	7.867	2	Good
P2	0.200	0.667	0.600	1.000	0.800	0.625	1.000	0.571	0.800	6.263	17	Average
P3	0.400	0.667	0.600	1.000	0.800	0.875	0.750	0.429	1.000	6.520	14	Average
P4	0.400	0.667	0.600	1.000	0.600	0.625	0.750	0.429	1.000	6.070	18	Average
P5	0.200	0.667	0.800	1.000	1.000	0.625	0.750	0.714	0.800	6.556	13	Average
P6	0.200	0.667	0.800	1.000	0.800	0.875	0.750	0.857	1.000	6.949	10	Average
P7	0.200	0.000	1.000	1.000	1.000	0.875	1.000	1.000	0.800	6.875	11	Average
P8	0.200	0.000	1.000	1.000	1.000	1.000	1.000	0.857	1.000	7.057	9	Good
P9	0.200	0.000	0.800	1.000	1.000	0.500	0.750	0.429	0.400	5.079	22	Average
P10	0.200	0.333	0.400	0.667	0.800	0.375	0.750	0.143	0.400	4.068	25	Poor
P11	0.400	0.667	0.400	0.667	0.400	0.375	1.000	0.000	0.600	4.508	24	Average
P12	0.400	0.333	0.200	0.667	0.200	0.125	0.750	0.143	0.200	3.018	27	Poor
P13	0.200	0.667	0.200	0.667	0.400	0.125	0.750	0.000	0.200	3.208	26	Poor
P14	0.200	0.000	0.800	1.000	0.800	0.625	1.000	0.714	0.400	5.539	20	Average
P15	0.200	0.667	0.800	1.000	0.800	0.625	1.000	0.714	0.600	6.406	16	Average
P16	0.400	0.333	1.000	1.000	1.000	0.750	1.000	0.857	0.800	7.140	7	Good
P17	0.400	0.333	1.000	1.000	1.000	0.750	1.000	0.857	0.800	7.140	7	Good
P18	0.400	1.000	1.000	1.000	1.000	0.875	1.000	1.000	1.000	8.275	1	Good
P19	0.200	0.000	1.000	1.000	1.000	0.750	1.000	0.714	1.000	6.664	12	Average
P20	0.200	0.333	0.800	1.000	0.800	0.625	1.000	0.571	0.600	5.930	19	Average
P21	0.200	0.667	1.000	1.000	1.000	0.875	1.000	0.857	0.800	7.399	6	Good
P22	0.200	0.000	0.800	1.000	1.000	0.625	0.750	0.571	0.200	5.146	21	Average
P23	0.200	0.333	1.000	1.000	0.800	0.625	1.000	0.857	0.600	6.415	15	Average
P24	0.200	0.333	1.000	1.000	1.000	1.000	1.000	1.000	1.000	7.533	4	Good
P25	0.200	0.000	0.600	0.667	0.600	0.625	1.000	0.571	0.600	4.863	23	Poor
P26	0.200	0.333	1.000	1.000	1.000	0.875	1.000	1.000	1.000	7.408	5	Good

). No policy reached the Excellent range, which informed subsequent weak-link diagnosis and differentiated optimization-path identification by rating category.

4.3.2. Visualization with PMC Surfaces

1. Policies Rated “Good”

Eight policies were rated “Good.” Their PMC surfaces showed a high-plateau pattern with localized depressions (Figure 14). At the primary-variable level, this group scored consistently high on policy target groups (X3), policy perspective (X4), policy focus (X5), and content quality assessment (X7), indicating relatively comprehensive specification of governance actors and objects and clearer articulation of a goals-measures-monitoring structure. Scores for policy instruments (X6), sustainable-governance assessment (X8), and policy safeguards (X9) were also generally high, reflecting stronger provision of instrument mixes and institutional supports, including elements relevant to sustainability such as intergenerational transmission, community participation, authenticity protection, risk control, digital safeguarding, and monitoring and evaluation. Within-group differentiation was driven mainly by policy legal status (X1) and policy time horizon and validity (X2). X1 remained comparatively low, suggesting that several “Good” policies were issued in lower-rigidity forms, which limits normative enforceability. X2 varied more substantially, indicating uneven temporal institutionalization (e.g., validity periods, phased tasks, and revision arrangements). Overall, the binding force and long-term operating design, rather than agenda coverage, accounted for the principal weaknesses within the “Good” group. Based on the PMC surface morphology, the eight “Good” policies were classified into three types:

(1) Single-depression type (P18). P18 exhibited the flattest surface, with its primary depression concentrated in X1; the remaining dimensions approached a near-ceiling plateau. This profile indicates strong coverage of core elements (actors, instruments, safeguards, and evaluative arrangements), while the lower X1 suggests constraints associated with the document’s hierarchical and legal-institutional status.

(2) Low X1 with moderate X2 type (P1, P21, P26; P24 as a weaker variant). P1, P21, and P26 showed a pronounced trough in X1, a moderate level in X2, and high plateaus elsewhere. This configuration reflects relatively complete governance frameworks and robust instrument-and-safeguard provisions, but only moderate temporal institutionalization (e.g., phased targets and rolling revision arrangements). P24 displayed a similar surface shape but with a lower X2, indicating larger gaps in time-bound design (validity periods, staged tasks, and rolling adjustment provisions).

(3) Key-dimension deficit or dual-depression type (P8, P16, P17). P8 was characterized by a deep depression in X2, indicating limited temporal-loop specification (time horizons, staged tasks, and evaluation-revision arrangements). P16 and P17 showed low X1 and X2, with an additional secondary depression in policy instruments (X6), suggesting comparatively insufficient or imbalanced instrument mixes.

Overall, the “Good” group showed relatively high scores in multi-actor coverage, agenda completeness, embedding of sustainability-relevant elements, and enabling safeguards, indicating a baseline capacity for long-term ICH governance. Their common constraints were concentrated in X1 and X2, and some policies also exhibited weaker X6. Accordingly, improvement priorities were to strengthen temporal institutionalization (X2), reinforce instrument configurations (X6), and enhance normative and institutional codification (X1) to improve cross-cycle enforceability and coordination stability.

2. Policies Rated “Average”

The PMC surfaces of the “Average” group typically showed a high plateau with either a single-corner collapse or a steep slope (Figure 15). Most policies maintained relatively complete coverage of governance frameworks and task design, but sharp depressions in a few key dimensions reduced overall performance to a moderate level. These depressions primarily reflected two deficiencies in sustainability-oriented governance design: limited compositional complexity in X1, with many texts confined to a single document form (e.g., plans or notices) rather than combining stronger legal normative authority with operationally binding provisions; and weak temporal institutionalization in X2, indicated by missing validity periods, phased timelines, and rolling revision mechanisms. Consequently, these policies functioned more as stage-based mobilization instruments than as institutionalized closed-loop arrangements that support sustained learning and corrective adjustment.

Further classification by surface morphology identified three types, which guided a decompositional reading of the “Average” policies.

(1) Single-corner collapse (P5, P6, P14, P15, P19, P23). These surfaces were largely flat but showed a pronounced trough at the same corner, indicating broadly systematic design with one critical weak link. The policies generally covered supply-demand and regulatory perspectives, inventory-based governance and transmission capacity building, and instrument mixes. However, they scored low on X2 and X9, consistent with absent provisions for implementation deadlines, evaluation feedback, and performance-linked arrangements. Because core enabling mechanisms for sustainable governance (organizational coordination, resource guarantees, accountability, and data support) are concentrated in X9, the trough reflects a structural constraint on sustained implementation, rather than a marginal omission.

(2) Steep-slope imbalance (P7, P9, P20, P22). These surfaces increased sharply from one side to a plateau on the other, indicating pronounced crossdimensional disparities. The low side was concentrated in X2 and X9. Specifically, some policies listed tasks and measures but lacked rolling updates, milestone setting, and monitoring-evaluation-revision arrangements, weakening the formation of a feedback loop. Others included resource inputs or action-oriented instruments but provided limited specification of cross-departmental coordination, statistical/data support, and performance incentives.

(3) Multi-undulation or saddle type (P2, P3, P4, P11). These surfaces showed multiple shallow troughs, indicating relatively broad coverage but weaker specification across several dimensions. The depressions were especially salient in the sustainability-based X8, which captures provisions on intergenerational transmission, community co-governance, authenticity and integrity, risk-bounded utilization, equity and inclusion, digital openness/sharing, and monitoring and evaluation linked to adaptive adjustment. Where X8-related positions declined, the texts provided fewer operational clauses for these elements, and the design appeared less integrated with mechanisms such as participatory arrangements, explicit risk boundaries, and dynamic monitoring.

To illustrate this structural inconsistency, P3 serves as a representative saddle-type case. In the PMC surface, P3 performs strongly in X4 (Policy Perspective), reflecting its broad framing of ICH governance across safeguarding, transmission, utilization, and institutional coordination. The text articulates multi-level objectives, aligns with national cultural strategy, and references cross-sector collaboration, indicating a comprehensive governance vision. However, its relatively low score in X8 (Sustainable Governance) suggests that this perspective is not fully translated into operational sustainability mechanisms. Although the policy emphasizes inheritance and development in principle, it provides limited specification of intergenerational incentive structures, participatory co-governance procedures, authenticity risk boundaries in market utilization, or formalized monitoring-evaluation-feedback cycles. In other words, the document excels in normative orientation but under-specifies the institutional arrangements that would embed sustainability into routine administrative practice. The resulting “saddle” morphology, thus, reflects a gap between discursive comprehensiveness and mechanism-level codification, rather than fragmented agenda coverage.

Overall, the “Average” group’s main deficit was not agenda coverage but weaker provisions for X1, X2, X8 and X9. Many texts lacked iterable timelines, milestone management, and rolling revision, and provided limited clauses on organizational coordination, resource guarantees, performance incentives, and supervisory feedback, which constrained operational, evaluable, and correctable implementation. Accordingly, optimization priorities were: strengthen safeguards and accountability arrangements (X9), institutionalize temporal governance (X2), increase normative codification and compositional complexity (X1), and improve the fit between instrument provision and implementation pathways (X6 and X7).

3. Policies Rated “Poor”

The “Poor” group (P10, P12, P13, and P25) displayed PMC surfaces with pronounced depressions and steep slopes (Figure 16), indicating highly uneven element configurations. Some policies showed partial operationalization in policy instruments and task specification and content evaluation. However, all four scored low on X1 and X2, and exhibited concentrated deficits in X8 and X9. Specifically, X8 components such as intergenerational transmission, community co-governance, authenticity protection, risk-bounded utilization, equity and inclusion, digital governance, and adaptive adjustment mechanisms were weakly embedded. This configuration separated actionable measures from the institutional conditions required for durable implementation, including clear responsibility allocation, monitoring and evaluation, feedback, and corrective revision.

At the individual-policy level, P12 and P13 showed the deepest surface depressions, with low scores concentrated in X6, X8 and X9. Both also scored lower in X3. This profile indicates reliance on principle-based statements and limited action arrangements, without a closed-loop structure for cross-sector coordination, societal participation, and sustained monitoring and evaluation. By contrast, P10 and P25 exhibited localized troughs that lowered overall performance, with deficits concentrated in X2, X8, and X9. Specifically, shortcomings included weak temporal design under X2, and limited institutionalization under X8 of mechanisms for authenticity protection, bounded utilization and commercialization risk control, rules for digital sharing, and dynamic monitoring with corrective adjustment. P27 retained relatively high scores in some dimensions but remained weak in X1 and X2, reflecting unclear policy status and time-governance design. This configuration combined comparatively strong measures with limited institutional codification, and was associated with its placement in the low-rating category.

Accordingly, capacity-building priorities for this group concentrated on four dimensions. First, strengthen policy status and temporal design (X1,X2) by tightening normative codification, specifying validity periods and phased timelines, and institutionalizing rolling revision. Second, operationalize sustainable-governance elements (X8) by codifying intergenerational transmission, community participation, authenticity protection, utilization risk control, and monitoring-and-evaluation requirements as executable clauses. Third, reinforce policy guarantees (X9) through explicit provisions for coordination, resource support, capacity building, responsibility allocation, and performance-linked feedback. Fourth, calibrate instrument provision (X6) and implementation arrangements (X7) to connect policy measures with sustained execution and corrective adjustment.

5. Discussion

This study applied a sustainable-governance lens to systematically evaluate China’s national-level ICH safeguarding policies. It operationalized SDG-aligned requirements—safeguarding and transmission, inclusive participation, capacity building, responsible utilization, and governance resilience—by translating them into observable mechanisms and explicit evaluative dimensions, thereby rendering sustainability empirically assessable within policy texts. Methodologically, the analysis established an integrated chain from agenda mapping to mechanism diagnosis. Text mining identified agenda structures, inter-topic relations, and stage-wise evolution, while the PMC index quantified policy-element configurations and PMC surface visualizations localized weak links and cross-policy differences. This design distinguished topic salience from institutional provisions that enable policies to be implementable, monitorable, and revisable. Accordingly, topic flows indicate shifts in agenda emphasis, but they do not mechanically imply improvements in policy design quality. PMC scores increase only when these thematic emphases are translated into codified mechanisms (e.g., time-bound pathways, monitoring-evaluation routines, feedback-driven revision, and enforceable safeguards), which explains why some later-stage policies remain “Average” despite agenda upgrading.

Empirically, the 26 policies yielded a mean PMC score of 6.150 (8 “Good”, 14 “Average”, 4 “Poor”), with none reaching the highest rating range. Under the original PMC grading logic, “Excellent” corresponds to very high internal consistency (typically ≥9), whereas scores in the mid-6 range are commonly classified as “Average”, indicating moderate completeness rather than an exceptional benchmark [93]. To contextualize our value, PMC-based evaluations in other Chinese policy domains often report similar mid-6 averages—such as 6.59 for green development policies, 6.67 for high-tech industry policies, and 6.26 for shipping decarbonization policies—suggesting that our result is broadly within the typical performance band observed in comparable PMC applications, albeit with clear room for improvement in sustainability-critical mechanisms [86,96,97]. The findings indicate a relatively stable institutional framework but incomplete embedding of sustainable-governance mechanisms, especially those supporting long-term orientation, adaptive adjustment, and closed-loop governance. Overall, the empirical evidence provides a direct check on the hypotheses proposed in Section 3.4. Overall, the stage-wise comparison supports H1, indicating an upward tendency in overall PMC performance in the later period, although the system-level mean remains at a moderate (“Average”) level rather than reaching an “Excellent” tier. The findings also lend support to H2, as later-stage policies increasingly incorporate sustainability-relevant governance arrangements and implementation provisions, consistent with the observed strengthening of mechanism embedding relative to earlier stages. In line with H3, temporal institutionalization remains a recurrent bottleneck: explicit milestone governance, rolling revision triggers, and fully operational time-bound pathways are still unevenly codified across national-level texts, which constrains the formation of a robust closed-loop governance architecture. This hypothesis-oriented interpretation clarifies which patterns were theoretically expected and helps position the “average” policy quality result as a structured diagnosis of mechanism incompleteness rather than a purely descriptive outcome.

Conceptually, the results differentiate policy coverage from governance capacity: sustainability performance depends on whether policies institutionalize a coherent chain linking objectives to implementation, accountability, and iterative learning. This distinction resonates with prior work arguing that policy capacity is determined less by agenda breadth than by the presence of enforceable instruments [98], accountability arrangements, and feedback mechanisms in the policy processes [99,100]. Where our study extends this line of research is that we operationalize the “chain” as observable, text-based mechanisms and assess their completeness systematically at scale. Practically, the framework functions as a reusable ex ante diagnostic tool to identify bottlenecks, compare designs over time, and target mechanism-level improvements that strengthen long-term governability, monitorability, and adaptability in national ICH safeguarding. This practical orientation is consistent with policy design and ex ante evaluation scholarship [101], which emphasizes diagnosing instrument mixes and implementation conditions early to reduce design-implementation gaps and improve long-term policy effectiveness [102,103].

Building on prior ICH policy studies that map thematic emphases or instrument repertoires [104], our evidence chain links agenda structure to instrument configuration and then to governance-capacity diagnosis, thereby allowing a mechanism-level interpretation rather than an inference from coverage alone. Topic identification indicates that China’s national ICH policy discourse is anchored in an institutionalized backbone—inventory governance and key-item safeguarding—co-occurring with implementation arrangements, standardized identification, and participation-oriented communication, and forming an architecture spanning core frameworks, regulatory execution, social mobilization, and resource support. Crucially, the PMC-based diagnosis does not merely confirm this breadth; it also points out the areas where design integrity has been compromised, indicating that the defects are concentrated in those clauses that are crucial for sustainability. These clauses enable monitoring, assessment, accountability and corrective actions based on feedback, which were not mentioned in previous studies. In practice, policies that are instrument-rich in outreach, project construction, industrialization, or culture-tourism integration more often exhibit deficits in sustainability-critical provisions, including authenticity and integrity safeguards [105], long-term security for transmission actors, rights and responsibilities for community participation, and risk and negative-externality governance [106]. Put differently, “usable” provisions are frequently specified, whereas the constraints that make utilization sustainable remain comparatively under-codified. From an SDG-alignment perspective, this pattern indicates that texts can quickly accommodate development-oriented agendas, yet without baseline constraints, risk governance rules, and feedback-driven revision mechanisms, they are unlikely to accumulate durable capacity under cross-cycle uncertainty. Accordingly, cultural sustainability hinges not on agenda breadth, but on whether policy design institutionalizes enforceable boundaries, accountability mechanisms, and iterative learning that stabilize safeguarding outcomes while coordinating inclusive development and responsible utilization.

Beyond mapping agendas and instrument mixes, this study identified structural constraints in the sustainable-governance mechanisms embedded in the sampled policies. The dominant weaknesses were not attributable to limited topic coverage; rather, they reflected insufficient closed-loop institutionalization and weak cross-cycle constraint capacity. First, long-term orientation and time governance were comparatively under-specified: stage-based deployments were not consistently translated into iterable timelines, milestone governance, and rolling revision arrangements, reducing implementation stability and adaptability across policy cycles. Second, the governance toolkit shows a systematic bias towards project-based safeguards and administrative management [107], with limited connection to systematic mechanisms such as institutionalized community participation [108], clear risk boundary rules, and dynamic monitoring. This, to some extent, echoes previous research. Third, safeguarding, accountability, and institutional-complexity provisions remained incomplete: codification of inter-organizational coordination, resource pooling, performance-linked incentives, and supervisory feedback was uneven, leaving the chain required for policies to be implementable, evaluable, and correctable only partially realized. Collectively, these deficits explain why no sampled policy reached the highest evaluation tier.

From the SDG governance logic, these weaknesses converge on a single deficit: the policy system lacks institutionalized implementation mechanisms that convert value-oriented objectives into arrangements that are measurable, accountable, and iteratively revisable. Put differently, the architecture specifies what to do more fully than how to sustain action over time, who bears responsibility, and how implementation is to be monitored, evaluated, and corrected. Because these mechanisms determine whether safeguarding and transmission can be consolidated into durable governance capacity, they are also pivotal for aligning ICH governance with SDG priorities, including inclusive development, decent livelihoods, and responsible utilization.

On this basis, the study advances a set of sustainable-governance strengthening pathways for optimizing China’s national-level ICH policy design. Oriented toward SDG requirements for safeguarding and transmission, inclusive participation, capacity building, responsible utilization, and governance resilience, these pathways target the mechanism-level institutional gaps identified in the empirical diagnosis.

(1) Institutionalize time to stabilize the tempo of sustainable governance. Policy design should couple long-term objectives with staged tasks through traceable annual plans, phased milestones, and rolling revision mechanisms, and should embed evaluation feedback and corrective-action procedures. This shifts policy from episodic deployment to continuous implementation with adaptive adjustment, strengthening cross-cycle stability and responsiveness. Policy texts should also codify objective-expiry checkpoints by specifying annual milestones, responsible actors, and deliverables, and by defining triggers for cross-year rolling evaluations and clearly delegated amendment authority, ensuring an operational temporal governance loop rather than mere directional guidance.

(2) Codify baseline rules to resolve the structural tension between utilization and safeguarding. For provisions promoting revitalized use, dissemination, and industrial integration, policy texts should be paired with enforceable safeguards for authenticity and integrity, and should specify development boundaries, risk-identification procedures, and negative-externality control rules. They should further institutionalize eligibility criteria, process monitoring, and exit/suspension mechanisms, thereby converting “permitted use” into sustainable use and preventing high-ambition design with weak constraints. In addition, a tiered risk-management regime and negative-list system should be introduced to subject commercial exploitation, tourism performances, and digital dissemination to risk grading and carrying capacity assessment, supported by ex ante access control, in-process monitoring, and ex post accountability, forming a utilization framework that is usable yet bounded.

(3) Embed communities and bearers in long-term safeguards and rights-and-responsibilities arrangements. Policies should institutionalize participation by heritage bearers and communities by specifying participation procedures, delineating rights-responsibilities boundaries, and defining principles for benefit sharing. In parallel, they should establish long-term provision mechanisms for training support, social protection, and public cultural services, while strengthening the governance capacity of grassroots organizations and multi-departmental coordination. Together, these measures form a transmission-governance support system centered on the continuity of primary actors. Concretely, policy texts can add stability clauses for bearers and communities, including sustained training and insurance support, rules for benefit sharing and intellectual property protection, and formal pathways for consultation, grievance submission, and remedy. They should also clarify the coordinating roles of grassroots institutions and industry associations, thereby enhancing actor resilience and governance equity.

(4) Operationalize closed-loop governance through data infrastructure and performance-linked instruments. Policies should strengthen provisions for cross-departmental coordination, resource guarantees, accountability enforcement, and information disclosure by establishing a unified system for data collection and monitoring and evaluation. Evaluation results should then be institutionally linked to budget allocation, program adjustment, and oversight and accountability, so that governance shifts from “having measures” to being evaluable, comparable, and correctable, thereby driving overall convergence toward higher policy quality. In addition, policy texts can standardize data definitions and indicator repositories, expand mechanisms for interdepartmental data sharing and public disclosure, and introduce interfaces for third-party evaluation and societal oversight. These arrangements enhance implementation transparency and traceable accountability, and ultimately strengthen the policy system’s adaptability and resilience in pursuit of long-term objectives.

6. Conclusions

Situated within the SDGs’ governance agenda of heritage safeguarding, inclusive participation, capacity building, responsible use, and resilience, this study conducted a systematic test of the sustainable-governance capacity embedded in China’s national-level ICH policy texts. Three conclusions stand out. First, China’s national ICH agenda is organized around a clearly institutionalized backbone centered on inventory-based governance and key-item safeguarding, and it has progressively extended toward standardization, implementation coordination, and public communication and participation. This trajectory indicates an intensifying chain of institutionalized identification, standardized execution, and social mobilization. Second, overall policy text quality remained moderate, and no sampled document reached an “excellent” range, suggesting that while a baseline of institutionalization has been established, high-level sustainable governance has not yet been consolidated. Third, the main obstacle to further upgrading policy quality lies less in agenda expansion than in the political-institutional conditions under which national ICH policies are produced and implemented. National-level texts often need to remain broad enough to accommodate highly heterogeneous heritage types and local governance capacities, which encourages a preference for principle-based guidance and programmatic mobilization over tightly specified, time-bound pathways. At the same time, ICH governance is inherently cross-sectoral (culture, education, tourism, finance, social security, digital governance), so stronger long-term arrangements and closed-loop provisions would require clearer inter-agency responsibility allocation, stable multi-year budgeting, and enforceable accountability—elements that are institutionally costly and politically sensitive to codify at the national level. As a result, policies may more readily articulate “what to promote” (safeguarding actions, utilization, dissemination) than “how to sustain and correct” action over multiple cycles (milestones, revision triggers, feedback-linked adjustment, and enforceable safeguards). This structural tendency helps explain why sustainability-critical provisions—especially those that lock in long-term incentives for bearers and communities, define risk boundaries for utilization, and institutionalize monitoring-evaluation-revision loops—are comparatively under-specified, leaving the system closer to a project-style safeguarding logic than to a fully institutionalized, iteratively improvable governance regime.

This study makes a twofold contribution. Theoretically, it integrates topic identification, relational visualization, and index-based diagnosis into a single analytical chain, moving policy research beyond agenda summarization toward structural-element assessment and weak-link localization. In doing so, it converts “sustainable governance” from a normative slogan into a measurable, comparable, and reusable analytical framework. This approach uses textual evidence to explain how, and why, policy systems do or do not crystallize long-term orientation, adaptability, and closed-loop governance capacity, thereby providing methodological support for advancing SDG-aligned priorities in the heritage field, including inclusive participation, capacity building, responsible use, and resilience enhancement. Practically, the indicator-based toolkit developed here enables policy self-auditing, iterative refinement, and cross-period comparison. It helps decision-makers identify institutional bottlenecks and structural deficits, while offering actionable evidence for strengthening monitoring and evaluation, feedback-driven revision, clause configuration, and closed-loop governance performance. Crucially, the study’s sustainability value does not lie in directly measuring SDG outcome indicators; rather, it provides a reproducible governance capacity diagnostic instrument that facilitates dynamic correction and continuous improvement during long-term implementation, thereby increasing the institutional feasibility of cultural sustainability.

This study has several limitations. First, the sample focused primarily on national-level policy texts and did not systematically incorporate subnational policies or implementation bylaws, which may understate regional heterogeneity and the diversity of grassroots execution mechanisms. Second, the scoring relied on indicator specifications and explicit textual statements, and therefore cannot fully capture implementation effects or implicit institutional arrangements. Third, topic modeling and stage delineation were constrained by corpus size and temporal distribution. Future research should extend the analysis to provincial and municipal policies and project-level implementation data, and triangulate policy text evidence with external sources such as budget allocations, performance evaluations, and societal feedback to build an integrated, linked evaluation system.