[Purpose/Significance] To explore the influence mechanism of Chinese traditional cultural elements on the user experience of video games, provide theoretical references and practical strategies for the integration of traditional culture into video game design, and further explore the effective paths for the contemporary communication of traditional culture among young people. [Process/Method] Taking Black Myth: Wukong as the research object, this study integrated theories of the technical communication of the traditional culture, symbolic interaction, and user experience. An experimental study was conducted using eye-tracking technology, where participants completed the prologue and combat events of the game. Eye-tracking metrics, such as the fixation count and total fixation duration in four areas of interest (characters, landscapes, combat, and costumes) were collected. Correlation and regression analyses were performed by combining data from a traditional cultural element perception questionnaire and the Game Experience Questionnaire (GEQ). [Result/Conclusion] Visual symbols of characters and costumes dominate the sensory experience. Dynamic symbols of characters, landscapes, and combat drive the interactive experience, while the cultural decoding of combat symbols triggers the value experience. Fixation metrics related to costumes and combat have significant explanatory power for users’ perceived competence and jointly form the basis of the immersive experience. The study also reveals the underlying reasons for the lack of a significant correlation among flow, emotional dimensions, and eye-tracking metrics. The findings provide theoretical references and practical strategies for integrating traditional Chinese cultural elements into video game design.

[Purpose/Significance] To overcome the limitations of traditional static topic analysis, which insufficiently considers dynamics and industrial relevance when depicting the evolution of emerging technologies, and to address the survivorship bias caused by retrospective analyses based on global data, this study integrates a three-dimensional perspective of “industrial chain-technology topic-dynamic evolution” to propose a method for identifying and analyzing dynamic emerging technology topics oriented toward industrial chains. Through an empirical analysis of patent data from the Beijing hydrogen energy industry (2010–2024), this paper verifies the effectiveness of a forward-looking dynamic technology topic identification framework, offering theoretical and practical references for policy-making, resource allocation, and cross-industry method transfer. [Method/Process] This study develops a dynamic emerging technology topic identification framework that integrates an industrial chain perspective, comprising topic identification and optimization, industrial chain construction, and the mapping of topics to chain segments. The research innovatively introduces two dynamic observation methods—forward-looking (ex-ante) and backward-looking (ex-post) perspectives—corresponding to evolutionary tuning and global tuning strategies, respectively. Using a topic similarity network, this study constructs intertemporal evolutionary paths and utilizes river charts to visualize the temporal evolution characteristics of technological topics. A three-dimensional analysis is conducted: evolutionary pattern identification, stage division, and key topic identification. [Result/Conclusion] Empirical findings reveal that the evolution of Beijing hydrogen energy technology has experienced three phases: technology dispersion, technology integration, and system integration. Two typical evolutionary patterns, continuous and disruptive evolution, are identified. Key topics such as “Chemical Separation and Reforming Technology” and “Hydrogen Energy System Integration Technology” form the core technological innovation pathways, while topics like “Hydrogen Energy Drone System Technology” demonstrate leapfrogging breakthrough characteristics. Technological topics exhibit a diversified distribution of upstream hydrogen production technologies, a relatively concentrated midstream storage and transportation sector, and a dense downstream application segment, reflecting the industry’s trend toward integration and large-scale development. The constructed framework, which combines forward-looking and backward-looking perspectives, has significant advantages in avoiding survivorship bias and capturing the uncertainties of technological evolution, providing a systematic and reusable analytical tool for the emerging industry technology evolution.

[Purpose/Significance] Technology convergence continuously generates new technological trajectories through cross-domain knowledge recombination and has become a critical mechanism driving industrial upgrading and technological paradigm shifts. Systematically identifying convergent technology combinations with high development potential is of substantial theoretical and practical significance for forward-looking technology opportunity discovery and the strategic allocation of innovation resources. [Method/Process] Based on the fundamental characteristics of technology convergence, this study develops a multidimensional evaluation framework from the perspectives of technology fusion and technology novelty. Technology combinations constructed from IPC co-occurrence at the patent level are used as the analytical units. An entropy-based weighting method is applied to determine the relative importance of the core dimensions, forming a quantitative model for assessing convergence potential. An empirical analysis is conducted using patents in the autonomous driving domain. [Result/Conclusion] The findings reveal a clear synergistic effect between technology fusion and technology novelty in the evolutionary process of convergent technology combinations. The integrated evaluation results effectively differentiate technology combinations across different potential levels, with high-potential convergent technology combinations exhibiting stronger technological diffusion capabilities.

[Purpose/Significance] In response to the key issues such as weak reproducibility and difficulty in multi-modal response assessment in the current evaluation of government intelligent Q&A systems, this paper, based on the benchmark testing concept, integrates large language models and machine learning technologies to construct a set of standardized evaluation question sets and an automated assessment framework applicable to government affairs scenarios, in order to support the scientific quantification and continuous optimization of system performance. [Method/Process] By systematically collecting and cleaning government affairs hot issues, a high-quality initial question set was formed; using an automated process to obtain the real responses of representative government intelligent Q&A systems and integrating the semantic understanding of large language models with traditional feature extraction methods, a set of answer keywords that can be used for automated judgment was generated; finally, a standardized evaluation question bank covering government affairs topics in multiple fields was constructed, and batch testing and quantitative evaluation were conducted on provincial-level government intelligent Q&A systems in China. [Result/Conclusion] The experimental results show that the automated evaluation method proposed in this paper has a significant consistency with the manual evaluation results, with a rank correlation coefficient of 0.77. Based on 10 rounds of actual tests and regression extrapolation, the dispersion of evaluation results can be kept within 5% over up to 27 rounds, demonstrating good robustness and reproducibility. This benchmark testing method has strong universality and scalability, and can not only be used for horizontal comparison between systems but also effectively track the performance changes of systems during minor version iterations, providing a feasible evaluation tool and reference basis for improving the quality of online government question-answering services.

[Purpose/Significance] To solve problems such as semantic ambiguity and incomplete structure of user queries and improve the effectiveness of medical query intent recognition, an intent recognition method RRQ-LMIR based on LLM retrieval re-ranking and query optimization is proposed. [Method/Process] Firstly, a hybrid retrieval strategy is utilized to obtain candidate examples with semantics similar to the user’s original query from an external knowledge base. Then, structured prompt information is constructed to guide the LLM to re-rank the candidate results in terms of semantic relevance. Subsequently, based on the re-ranking results, the LLM is utilized to generate an optimized query with clearer semantics and a more complete structure. Finally, a fine-tuned deep learning model is adopted to classify the intentions of the initial query and the optimized query, and the final prediction result is obtained through an equal-weight fusion strategy. [Result/Conclusion] The experimental results show that the F1-score of the RRQ-LMIR method for medical query intent recognition constructed in this paper reaches 84.30%, which is 3.26% and 22.7% higher than those of the current mainstream deep learning model RoBERTa-wwm-ext and the large language model GPT-5, respectively. The method proposed in this paper effectively solves problems such as semantic ambiguity and incomplete structure, and improves the performance of medical query intent recognition.

[Purpose/Significance] This study aims to enhance large language models’ (LLMs) capability for knowledge extraction through a multi-stage strategy, addressing performance bottlenecks in the specialized biomedical contexts, and providing a replicable path for knowledge organization, extraction, and discovery in pharmacological field. [Method/Process] A knowledge organization framework covering 12 entity types and 26 relation types was constructed, and domain-specific corpora were generated via prompt engineering. The proposed three-stage strategy (Multi-KE), integrating weak supervision, strong supervised fine-tuning, and reflective optimization, progressively improved the model’s extraction performance and interpretability. Multi-hop path discovery based on a knowledge graph was conducted to evaluate the model’s applicability in uncovering drug mechanisms and potential indications. [Result/Conclusion] On the PubMed dataset, Multi-KE achieved F1 scores of 82.20% for entity extraction and 65.54% for relation extraction, significantly outperforming general large models (GPT-4o, LLaMA3-8B) with improvements of 20.32%, 34.53% and 29.47%, 47.97% respectively, and performing only slightly lower than DeepSeek-R1 (by 2.44% and 3.48%). Meanwhile, it achieved substantial improvements of 22.54% and 35.49% over the domain-specific large model OpenBioLLM-70B. Furthermore, the ADE-Corpus-V2 dataset was used to validate the model’s generalization ability, and ablation studies confirmed the contribution of the three-stage strategy to performance improvement. Through data distribution and error analysis, the study also explored in depth the model’s performance and limitations in low-frequency entity recognition and complex relation extraction.Traceable extraction records and reflective optimization enhanced the model’s transparency and interpretability. Furthermore, through multi-hop path analysis, the proposed framework is validated for its effectiveness in identifying potential drug–disease associations and supporting the analysis of complex pharmacological mechanisms.

[Purpose/Significance] Focusing on the context of major emergencies, this study constructs a theoretical framework for the core concept of government information disclosure quality and follows a systematic scale development process to develop a Scale for Government Information Disclosure Quality in Major Emergencies, providing a reliable measurement tool for evaluating government information disclosure quality in such situations. [Method/Process] Based on the conceptual framework of government information disclosure quality in major emergencies established in previous research through coding analysis and expert consultation, an initial measurement scale was developed. Exploratory factor analysis(EFA) was employed to examine and refine the factor structure of the scale, and the dimensional structure derived from qualitative analysis was further optimized based on quantitative results. Confirmatory factor analysis(CFA) was subsequently conducted to evaluate the reliability, validity, and model fit of the revised scale. [Result/Conclusion] The final scale consists of 41 measurement items across five dimensions: quality of disclosure practices, quality of information presentation, quality of event reporting, quality of emergency response, and quality of public empowerment. The results indicate that the scale demonstrates good reliability and validity, providing a scientific and reliable instrument for assessing government information disclosure quality in the context of major emergencies. [Innovation/Value] The proposed scale integrates both general requirements of government information disclosure and the distinctive characteristics of major emergencies, particularly urgency, uncertainty, and public-oriented demands. It captures the specific nature of emergency information disclosure and highlights its role in supporting emergency response, protecting public interests, and strengthening public trust in government. Moreover, it addresses the limitations of existing measurement instruments that are often constrained by routine disclosure contexts or highly homogeneous indicators, thereby offering a more interpretable and broadly applicable tool for evaluating government information disclosure quality across different types of major emergencies. In addition, by operationalizing government information disclosure quality as a measurable construct, this study provides a foundation for future empirical research exploring its relationships with key variables such as public trust and social resilience. [Insufficiency/Improvement] Despite its contributions, this study has certain limitations. Major emergencies are characterized by strong contextual immediacy, meaning that measurement accuracy is higher when applied during or shortly after actual emergency events. However, the data in this study were collected under routine conditions, where respondents mainly relied on experience and recall, which may introduce response bias and limit the precision of measurement. Future research should further strengthen the theoretical framework by integrating perspectives from crisis management and information communication studies, and apply the scale to real emergency events to examine its sensitivity, adaptability, and robustness. This will help improve the instrument's applicability and enhance its practical value in evaluating government information disclosure quality across diverse emergency scenarios.

[Purpose/Significance] In the context of the information resources management(IRM) discipline, researchers frequently face challenges in selecting, applying, and innovating research methods due to the continuous expansion of data scales and the increasing complexity of knowledge systems. To address the fragmentation of research method knowledge and the insufficient intelligent service capabilities within the discipline, this study aims to develop a specialized intelligent question-answering(Q&A) service. By conforming to the new AI for Science(AI4S) paradigm, this research seeks to realize intelligent Q&A services for disciplinary research methods, thereby promoting continuous innovation and development in intelligent information analysis and knowledge system construction. [Method/Process] This study utilizes literature data from 21 journals within the IRM discipline published between 1995 and 2025 to construct a comprehensive research method knowledge set. The foundation of this knowledge set is a structured "paper-problem-method" framework built upon rigorous method entity extraction. To optimize entity extraction, various deep learning models and large language models(LLMs) were experimentally compared, identifying the SciBERT+BiLSTM+CRF architecture as the optimal approach for large-scale extraction tasks. Subsequently, a knowledge-enhanced large language model(KELLM) tailored specifically for IRM research methods was proposed. This model synergizes external knowledge enhancement, knowledge-constrained reasoning, and task-oriented prompt control. Finally, user experiments were conducted to assess the model's practical application effects. These experiments evaluated user satisfaction, result relevance, and result interpretability across varied scenarios, factoring in different user types and problem clarity conditions. [Result/Conclusion] The implementation of the KELLM successfully achieved the seamless integration, organization, and reasoning of research methods, providing a reliable methodological path for LLMempowered knowledge Q&A services in the IRM discipline. Evaluation results demonstrate that the KELLM significantly outperforms general LLMs in domain knowledge analysis and recommendation services. The model successfully conducts in-depth analysis and provides precise recommendations tailored to diverse user profiles and problem structures. Specifically, for domain users with clear questions, KELLM delivered highly precise method matching and decision support. For domain users with vague requirements, the model effectively broadened research ideas and provided heuristic guidance. Furthermore, for out-of-domain users, the KELLM demonstrated strong knowledge generalization, facilitating cross-disciplinary knowledge navigation and effectively meeting the demands of multiple Q&A scenarios. [Innovation/Value] A primary innovation of this study is the construction of an IRM research method knowledge set based on a ternary "paper-problem-method" structure, which provides a high-quality data foundation that significantly reduces the hallucination problems typically associated with general LLMs. By relying on external domain knowledge constraints for retrieval and reasoning, the proposed KELLM is more accurate and transparent than standard general models. This study not only provides a methodological reference for AI-driven knowledge services and academic innovation within the discipline under the AI4S paradigm but also offers valuable insights for constructing intelligent agents in related fields. Ultimately, it highlights that the core of LLM-driven knowledge services relies on high-quality underlying data and a robust internal knowledge logic architecture rather than mere engineering techniques. [Insufficent/Improvment] Despite its contributions, this study has limitations that present opportunities for future improvement. First, the current research method knowledge set relies exclusively on IRM journals and does not encompass cross-disciplinary data, such as computer science literature; integrating such interdisciplinary data will serve as a key expansion target. Second, there remains room for improvement in the precision of the research method entity extraction task. Future studies should consider adopting multi-model fusion strategies to further enhance model performance and deliver more precise Q&A services. Lastly, the user evaluation was conducted as a preliminary exploratory experiment with a relatively small sample size. Future research will expand the sample scope to further validate and enhance the universality of the model.

[Purpose/Significance] With the development of agent technology, a new implementation path has emerged for intelligent library services. The emergence of agent skills greatly improves the functionality and business efficiency of intelligent library services. However, there is a lack of academic studies on existing skills and their application in libraries. This paper constructs a classification framework for library agent skills, discusses the development processes of these skills and their applications, and also provides ideas for the innovation of smart library services. [Method/Process] The research methods include literature review, content analysis, and expert interviews. Firstly, the literature on smart libraries and library agents is reviewed from 2021 to 2026. The literature is mainly searched in CNKI, Wanfang, ScienceDirect, and Google Scholar, and then supplementary searches are conducted in Baidu Scholar, PubMed, and IEEE Xplore. Core journal articles, master's and doctoral theses, industry research reports, and high-quality foreign studies within the last 5 years are analyzed. In the end, 33 representative references are obtained. According to the core business processes of library agent skills and library users' demands, the study systematically analyzes the functional structure and service forms of library agent skills, and then establishes an initial classification framework of library agent skills. Finally, in-depth interviews are conducted with experts. Experts' opinions are used to further confirm the scope of the framework and the logical division of categories, so that the framework can be more accurate and practicable.[Result/Conclusion] According to the research, the six main categories are: collection resource services, document circulation services, reference and consultation services, reading promotion services, learning support services, and smart space services. This study presents a skills development model based on the lifecycle process. The model shows the main stages of skills development and utilization. The foundational technologies of the skills mentioned above are also introduced, as well as their development models and practical applications in libraries. These results offer references for the design of library agents, intelligent service systems, and cognitive collaborative service models, thereby promoting the development of smart libraries with cognitive library architectures. [Innovation/Value] This paper introduces agent skills to the field of libraries and establishes a systematic classification for them. A development model is proposed based on lifecycle management theory, extending the theoretical scope of integrating smart libraries with generative artificial intelligence. These findings provide a basis for the design of library agents, the optimization of intelligent service systems, and the application of generative AI in libraries. [Insufficiency/Improvement] Empirical research has not yet been carried out, as the framework has not been tested in real library services or actual operations over a long period. More research is needed to evaluate the practical effects and operating mechanisms of agent skills. In the future, case studies will be conducted, and a prototype system will be constructed and tested with users to verify the practicality and generalizability of the skills framework. Further studies will also address complex business scenarios, multi-agent coordination, longterm memory mechanisms, and standardized ecosystem construction. This will help promote the standardization, continuous operation, and reuse of library agent skills.

[Purpose/Significance] To explore the conceptual framework and influencing mechanism of digital resilience of researchers in the context of large language models, in order to improve the academic efficiency of researchers and achieve deep integration between large language models and their academic research. [Method/Process] Based on the critical incident technique analysis method, this study refined and summarized data obtained from in-depth interviews with 28 scientific researchers from different disciplines, and constructed a research model for the conceptual framework of digital resilience of scientific researchers in the context of large language models, a conceptual framework research model of “psychological resilience-technical resilience-academic resilience” for researchers' digital resilience is proposed. [Result/Conclusion] The digital resilience of researchers in the context of large language models mainly includes three key nodes: self-adjustment, technological adaptation, and academic innovation. The influencing factors of researchers' digital resilience at different stages are analyzed. Based on the research results, strategies and suggestions have been proposed to emphasize the emotional intelligence of researchers, strengthen their technical resilience, and optimize their resilience and innovation.

[Purpose/Significance] Based on information ecosystem theory, this study adopts regression analysis, the GA-PSO-RF model, and the fsQCA method to systematically explore the influencing factors and configurational paths of AI deepfake video dissemination. It aims to provide useful references for formulating governance strategies to address the dissemination of deepfake videos and advancing the governance of the cyber ecosystem. [Method/Process] This study collected 338 deepfake video data points from the Bilibili and Douyin platforms, used regression analysis to identify influencing factors, employed the hybrid optimized random forest model to rank the importance of each influencing factor, screened key factors, and utilized fsQCA to conduct a configurational analysis of the key factors influencing the dissemination of deepfake videos. [Result/Conclusion] The study finds that communicator visibility, user age distribution, the number of video tags, visual communication technology level, audio-visual matching communication technology level, account technology content verticality, video duration, video title length, and video topic category significantly affect deepfake video dissemination. Two configurations were identified for each of high-level and non-high-level AI deepfake video dissemination, in which communicator visibility and account technology content verticality are the core conditions for accelerating video dissemination, and are also important focal points for implementing video dissemination governance strategies.

[Purpose/Significance] Policy ambiguity serves as a pivotal governance mechanism for governments to forge consensus and catalyze innovation in complex scenarios. Its significance is increasingly prominent against the backdrop of escalating global governance complexity and diversifying social interests. However, existing methods struggle with the large-scale, precise measurement of policy ambiguity. Developing a new measurement method to grasp its overall trends and evolutionary patterns is therefore of great significance for the demands of increasingly sophisticated national governance. [Method/Process] This study integrates expert knowledge with large language models to propose a novel policy ambiguity measurement method—ELKDL(Expert-Linguistics-Knowledge-Driven LLM). The method begins by extracting lexical-level ambiguity knowledge units via linguistic parsing. It then uses minimal expert annotations to drive the LLM in constructing a domain knowledge base. This knowledge base is subsequently used to guide the LLM in completing the ambiguity tagging of policy directive sentences, thereby enhancing the breadth and depth of the measurement. Concurrently, the study employed knowledge distillation to build high-precision models for directive identification(95.97%) and ambiguity tagging(92.01%), which were integrated with sentence segmentation and computational modules to form the complete measurement model. [Result/Conclusion] Expert validation demonstrates that the ELKDL method achieves a tagging accuracy exceeding 86.5%, significantly outperforming the best result from direct LLM judgment(64.44%). The study further verified the method's effectiveness in cross-domain and cross-hierarchical tasks, proving its capability for large-scale, high-quality, and automated measurement of policy ambiguity. This work not only provides an innovative methodology for policy research, formulation, and implementation but also offers a paradigm for integrating domain expert knowledge with large language models.

[Purpose/Significance] Education, science and technology, and talent policies serve as the cornerstone for supporting the innovation-driven development strategy and the workforce development strategy. Conducting quantitative evaluation research on these policies can provide a scientific basis for promoting the synergistic integration and high-quality development of the “trinity” of education, science and technology, and talent. [Method/Process] This study examines 344 education, science and technology, and talent policies, employing BERTopic and large language models (LLMs) to improve the traditional PMC index model. A PMC index model incorporating three-tier variables for education, science and technology, and talent was constructed. [Result/Conclusion] The study shows that China’s education, science and technology, and talent policies demonstrate strong overall consistency. However, further efforts are needed to clarify the timelines of policy implementation, enhance policy coordination mechanisms, and strengthen the synergy between central and local governments.

[Purpose/Significance] Accurately predicting the popularity of social information is critically important for government regulation, platform governance, and public decision-making, as viral misinformation can trigger social panic and disrupt public order. While deep learning methods have advanced this task, most operate under a homogeneous graph assumption that fails to capture the heterogeneous nature of real-world social networks. Even existing heterogeneous graph approaches typically encode high-level “topic” semantics as implicit edge attributes rather than explicit structural entities, severely limiting their capacity for fine-grained semantic reasoning across the complex, non-linear dependencies among users, texts, and topics. To address these fundamental gaps, this paper proposes a novel meta-path enhanced feature learning framework that explicitly materializes topics as first-class graph nodes and introduces an adaptive meta-path selection mechanism to fully exploit heterogeneous semantic associations. [Method/Process] The proposed method proceeds in three stages. First, a heterogeneous social graph is constructed with three node types, namely users (U), texts (V), and topics (T), where topics are extracted via LDA and modeled as independent nodes. Raw features from BERT, user influence metrics, and topic distributions are projected into a unified semantic space via type-specific transformations. An adaptive meta-path selection mechanism then automatically identifies predictive semantic patterns: it generates candidate paths through breadth-first search with type-constrained pruning, evaluates their importance using an attention mechanism that fuses structural and semantic path information, and employs a dual-regularization loss with L1 and entropy terms to dynamically reinforce high-contribution paths while eliminating redundant ones during training. Second, a two-stage feature enhancement strategy refines target node representations. Stage One applies a novel type-role-position three-factor attention mechanism that aggregates fine-grained semantics from all nodes along a meta-path, not merely the endpoints, by jointly modeling each node’s type, its path role, and its positional distance from the target. Stage Two introduces a cross-layer residual gating mechanism to combat over-smoothing in deep architectures, dynamically balancing shallow local features with deep global semantics to preserve local distinctiveness while integrating high-order structural information. Finally, enhanced features from multiple meta-paths are adaptively fused via path-level attention and fed into a Multi-Layer Perceptron for popularity prediction. [Result/Conclusion] Extensive experiments on a large-scale Sina Weibo dataset demonstrate the effectiveness of the proposed framework. Ablation studies validate the indispensable contribution of each component. Comparative results against eight representative baselines covering feature engineering, time series analysis, and state-of-the-art deep learning methods show that the proposed model achieves the best performance on both MSE and MAE metrics. Parameter sensitivity analysis is conducted to determine the optimal hyperparameter configuration, and topic perturbation experiments confirm the model’s robustness to suboptimal topic settings. Qualitative case studies further verify the model’s strong generalization ability across different popularity levels. [Innovation/Value] This work makes dual contributions. Theoretically, it provides a new heterogeneous graph learning paradigm by structurally elevating topics to explicit nodes and introducing an adaptive, data-driven meta-path selection mechanism that obviates manual design. The synergistic two-stage feature enhancement constitutes a novel methodological advance for capturing non-linear heterogeneous semantics. Practically, the proposed method can support public opinion monitoring and platform content management by enabling more accurate popularity prediction. [Insufficient/Improvement] Several limitations point to future work. The LDA-based topic modeling neglects deep contextual semantics, which can be improved with neural topic models. The current framework only leverages textual and structural features, omitting multimodal content pervasive in social media. Finally, cross-platform validation is needed to verify generalizability.

[Purpose/Significance] This paper constructs an interdisciplinarity measurement system based on Self-established Interdisciplinary Programs for Graduate Degrees (SIPs) in Chinese universities. On this basis, this paper aims to explore the overall construction characteristics of SIPs and specific combination patterns of the relevant Tier-2 Disciplinary Programs (T2Ps), and provide references for interdisciplinary construction and talent cultivation. [Method/Process] From the perspective of complex network analysis, several networks were constructed: the co-occurrence network and hypergraph of T2Ps; the coupling network of SIPs; and the three-mode network “SIPs-T2Ps-T1Ps (Tier-1 Disciplinary Programs)”. This paper uses the structural characteristics of the four types of networks and network visualization techniques to present the interdisciplinary features of SIPs, and constructs a measurement system that includes indicators such as cohesion and diversity. Comprehensive analyses were conducted by applying cohesion, diversity, and integration indicators to measure and analyze these networks and their clustering results. Among them, the co-occurrence network of T2Ps can reveal binary associations among T2Ps, while the hypergraph more accurately represents the interdisciplinary relationships among multiple T2Ps. The coupling network of SIPs, through community detection algorithms, helps obtain the overall research topics of SIPs. The “SIPs-T2Ps-T1Ps” three-mode network focuses on the three currently most popular types of SIPs and comparatively explores the knowledge integration characteristics formed by typical current SIPs from four aspects: the variety of T2Ps and T1Ps, the balance of T1Ps, and the disparity of T2Ps. [Result/Conclusion] T2Ps such as Computer Science & Technology and Management Science & Engineering lead prominently in both the breadth and strength of interdisciplinary engagement. The T2Ps involved in SIPs exhibit distinct integration features, such as medical-medicine fusion, medicine-biology integration, engineering-engineering crossover, and management interpenetration. The establishment of SIPs aligns with scientific research frontiers, national science & technology development strategies, and socio-economic needs, covering fields such as Artificial Intelligence (AI), Data Science, and Security. Security-focused SIPs involve the widest range of disciplines, evenly integrating specialized knowledge from different T2Ps, which also exhibit relatively pronounced differentiation among themselves. Finally, this paper discusses the development pathway for SIPs and their contributions to interdisciplinary talent cultivation, institutional restructuring, and organizational construction. [Innovation/Value] Building on previous studies, this paper adds the analytical perspectives of a hypergraph of T2Ps, a coupling network of SIPs, and a three-mode network of “SIPs-T2Ps-T1Ps”, thereby measuring more comprehensively and precisely the interdisciplinary characteristics exhibited by SIPs. In addition, this paper selects the latest catalog of SIPs and measures interdisciplinarity from the top-level design perspective of disciplinary construction, compensating for the shortcomings of previous interdisciplinarity measurement studies, which overemphasize measurement entities located in the mid- and lower reaches of the academic ecosystem, such as academic journals and funding project applications. [Insufficient/Improvement] In the measurement process, no conceptual, methodological, or instrumental distinctions were made regarding the interdisciplinary relationships among T2Ps, and no case analysis was conducted in conjunction with universities’ actual disciplinary construction, personnel composition, or organizational setup.

[Purpose/Significance] In the context of global technological competition and industrial transformation, technological competition has become the core battlefield for comprehensive national strength. China still faces a significant gap in certain key technology areas compared to international advanced levels. The increasingly complex and severe external technological environment exacerbates the "bottleneck" dilemma in specific technology fields. Against this backdrop, constructing a scientific and efficient "bottleneck" technology identification system has become an important task for safeguarding national technological sovereignty and promoting independent innovation in industry. [Method/Process] From the perspectives of technological dependence and technological blockade, this study integrates the relationship between technology pioneers and technology latecomers in the process of identifying bottleneck technologies and designs a framework for their identification. First, using patent data as the source, key core technologies are selected based on technological value and economic value, and the BERTopic topic model is employed to mine patent themes. Second, a patent citation network is constructed, and the technological impact is assessed using network vulnerability analysis methods. This assessment is complemented by a formula for measuring technological dependence based on patent citation relationships. Subsequently, the technological blockade index is measured using text similarity methods in conjunction with the Commerce Control List and patent data. Finally, the identified key bottleneck technologies and potential bottleneck technologies are selected based on the aforementioned indicators. [Result/Conclusion] China has several existing and potential bottleneck technologies in the chip domain, primarily concentrated in areas such as EUV lithography and photoresists, chemical vapor deposition, and electronic design automation. The results were validated against relevant industrial policies, confirming the reliability of this method. [Innovation/Value] First, by combining the dimensions of technology dependence of latecomers on pioneers and the blockade imposed by pioneers on latecomers, this study provides a more comprehensive view of the bottleneck technology issue. Technological dependence often leaves latecomers in a vulnerable position susceptible to blockade, while technological blockade further intensifies the latecomers' dependence on the pioneers' technologies. Second, the introduction of patent citation networks and complex network analysis methods allows for the construction of a technological dependence index, while the text similarity calculation between patent themes and Commerce Control List themes facilitates the development of a technological blockade index. [Insufficient/Improvment] It is challenging to effectively identify "sleeping beauty" patents that possess significant potential value but have not been widely cited. Future research could integrate the content of patent texts to build a comprehensive framework for evaluating explicit influence and potential disruption. Additionally, incorporating expert knowledge, supply chain information, and enterprise interviews into the data sources could establish a multi-dimensional analysis system encompassing policies, patents, and industrial chains, thereby enhancing the accuracy and comprehensiveness of the identification model.

[Purpose/Significance] Potential emerging technologies are prone to recognition difficulties due to their inherent fuzzy characteristics. This article aims to construct a recognition system that is adapted to their fuzzy characteristics, achieve effective recognition of potential emerging technologies, and provide support for the allocation of technological innovation resources and future industrial development. [Method/Process] Based on the fuzziness characteristics of potential emerging technologies and their manifestations, using patents, news, and reports as data sources, a recognition model is constructed from two feature dimensions: technical semantic fuzziness and application scenario fuzziness. Firstly, based on dynamic networks and fuzzy C-means clustering methods, cross-domain association strength and attribution uncertainty indicators are constructed to identify semantic fuzziness technology. Secondly, with the help of the BERTopic topic model, a fuzzy technology for identifying application scenarios based on technology application diversity and scenario transition indicators is constructed. Finally, assign weights are assigned to the comprehensive technical semantic ambiguity index and the comprehensive application scenario ambiguity index to identify potential emerging technologies. [Result/Conclusion] With the field of hydrogen fuel cells as the empirical object, based on the Pareto Principle, the top 20% percentile of the comprehensive score (i.e., comprehensive potential score ≥ 0.32) is set as the recognition threshold, and finally 5 potential emerging technologies are identified.

[Purpose/Significance] To overcome the limitations of traditional research that treats ambidextrous innovation as a monolithic construct and overlooks differences in its internal driving mechanisms, this paper, from a configurational perspective, aims to deconstruct and compare the antecedent configurations leading to high exploratory and high exploitative innovation performance, thereby revealing the asymmetric patterns between them and deepening the understanding of the ambidexterity paradox. [Method/Process] Based on the “Technology-Organization-Environment” (TOE) framework, which is, enhanced by integrating perspectives such as knowledge governance and “data empowerment” to deepen the conceptual understanding of boundary-spanning search and knowledge dynamic capabilities, this study employs fuzzy-set Qualitative Comparative Analysis (fsQCA) to empirically analyze data from 483 Chinese enterprises. By separately calibrating fuzzy sets for exploratory and exploitative innovation performance as outcome variables, it systematically examines the synergistic effects of boundary-spanning search, knowledge dynamic capabilities, and environmental complexity. [Result/Conclusion] The research identifies three pathways leading to high exploratory innovation performance and two pathways leading to high exploitative innovation performance. Comparative analysis reveals a fundamental asymmetry in their driving logics: high exploratory performance relies on the combination of boundary-spanning search and knowledge creation capability within complex environments, whereas high exploitative performance depends on the internal synergy of knowledge dynamic capabilities aligned with a stable environment. This study confirms the configurational roots of the inherent tension in ambidextrous innovation, providing important theoretical and practical insights for enterprises to contextually allocate resources and precisely manage these two types of innovation activities.

[Purpose/Significance] Against the backdrop of the profound shifts of the global innovation landscape and China's accelerated implementation of the innovation-driven development strategy, patents have evolved beyond legal instruments into core carriers of innovation and key indicators of regional competitiveness. While existing evaluation systems generally rely on a solid tripartite framework—comprising legal, technological, and economic values—they frequently struggle with dynamic adaptability in complex macro-strategic contexts. Traditional endogenous indicators fail to capture the forward-looking strategic orientation embedded in dynamic policy adjustments. Grounded in the Technological Innovation System (TIS) theory, this study argues that high-value patents should not only fulfill technical and legal criteria but also demonstrate alignment with macro-strategic needs. Therefore, this study aims to construct a quantitative analytical framework and an evaluation system for patent value that can dynamically adapt to macro contexts and integrate strategic perspectives. [Method/Process] Taking the Guangdong-Hong Kong-Macao Greater Bay Area (GBA) —characterized by cross-institutional, cross-regional, and cross-industrial collaborative innovation—as an empirical case, this study designs a research path of "knowledge discovery—rule mapping—dynamic weighting—empirical validation. " Specifically, in the knowledge discovery phase, the study combines the BERTopic model with rule-engine technology to mitigate the semantic sparsity problem of traditional bag-of-words models, effectively extracting latent strategic themes from exogenous policy texts. In the rule mapping phase, these themes are mapped into five supplementary categories of strategic value: industrial policy responsiveness, regional collaborative radiation, policy empowerment intensity, sustainable-development drive, and technical-rule dominance, which are further refined into ten secondary indicators. Finally, in the dynamic weighting phase, utilizing exogenous policy texts as the core benchmark, the study couples the rule engine with a time-decay coefficient to design a dynamic entropy-weight model. This process addresses static lag issues and constructs a comprehensive evaluation system encompassing legal, technological, economic, and strategic values. [Result/Conclusion] Multiple empirical and robustness tests based on the GBA patent samples indicate that the proposed evaluation system possesses demonstrates strong construct validity and criterion-related predictive validity. The empirical results verify the significant marginal explanatory power of strategic value as a supplementary dimension. Furthermore, heterogeneity analysis reveals distinct institutional constraints: strategic value exhibits a negative impact on patent transfer rates across all samples, but the magnitude is significantly stronger for non-enterprise entities (e. g., universities and research institutes) due to constraints regarding risks of state-owned asset losses and maturity levels, whereas enterprises tend to adopt defensive strategies to maintain core competitiveness. Overall, this measurement paradigm can identify core patent assets highly aligned with macro-strategic orientations, effectively broadening the measurement dimensions of traditional evaluation systems and enhancing their dynamic adaptability in complex policy contexts, thereby providing a quantitative decision-making reference for innovation-resource allocation. [Innovation/Value] This study makes three main contributions. First, theoretically, it breaks through the traditional micro-level evaluation perspective by establishing a structural mapping between macro-policy semantics and micro-patent evaluation. By introducing "Strategic Value" based on the TIS theory, it extends the evaluation dimension to macro-strategic alignment. Second, methodologically, it integrates the BERTopic model with a customized rule engine to achieve a rigorous and objective mapping from unstructured policy intentions to structured indicator weights. Third, in terms of evaluation logic, by adopting exogenous policy texts as the core benchmark for weighting rather than excessively relying on endogenous data, the study mitigates homological endogeneity bias. This creates a highly transparent, traceable, and dynamic measurement paradigm capable of objectively responding to macro-strategic guidance. [Insufficient/Improvement] Despite these theoretical and methodological advancements, certain limitations remain to be addressed in future research. First, the spatial generalizability of the evaluation system requires further verification. Future studies could apply this measurement paradigm to other strategic regions, such as the Yangtze River Delta and the Beijing-Tianjin-Hebei region, to confirm the generalizability and robustness of the evaluation logic through cross-regional horizontal comparisons. Second, due to the long lifecycle of patent value realization, current cross-sectional data cannot fully capture dynamic value trajectories. Subsequent research should establish long-term panel databases and employ methodologies such as survival analysis to conduct full-life-cycle tracking tracking of high-value patents' conversion rates, implementation rates, and ultimate economic benefits.

[Purpose/Significance] This study aims to provide methodological support for uncovering policy-relevant knowledge and illuminating the underlying mechanisms of policy processes, thereby advancing evidence-informed and scientific governmental decision-making. [Method/Process] Firstly, the conceptual connotations of policy citation are systematically reviewed, with an analysis of its practical manifestations and contextualized functions in policy documents, as well as its methodological framework as an analytical tool. Secondly, a policy process analysis framework is constructed from the perspective of policy citation, elucidating in detail its embedded pathways across the full policy lifecycle, including agenda setting and evolution, policy design and formulation, policy diffusion and implementation, and policy evaluation and optimization. Finally, the potential applications and future development directions of policy citation analysis are discussed. [Result/Conclusion] This study provides a novel methodological system for understanding the dynamic patterns and intrinsic characteristics of policy processes. It further expands the application scope of citation analysis methods in evidence-based policy-making, and supports high-quality, transparent, and sustainable policy formulation and governance practices.

[Purpose/Significance] This study explores the external and internal influencing mechanisms of user engagement in public digital cultural services (PDCS), which is of great theoretical and practical significance for enhancing user experience and improving PDCS effectiveness. [Method/Process] Based on the Stimulus-Organism-Response (S-O-R) theoretical framework, this study constructs a conceptual model of the influencing mechanisms of PDCS user engagement. It employs a fixed-mixed research method. Quantitative analysis is performed on questionnaire data via a structural equation model to test research hypotheses; qualitative analysis is performed on user interview data to interpret the findings and assist in the formulation of optimization strategies. [Result/Conclusion] The findings of this study indicate that in terms of the external influencing mechanism, expectation confirmation and trust beliefs have a significant positive impact on cognitive engagement. Basic psychological need satisfaction positively affects both cognitive and affective engagement, while privacy concerns have a significant negative impact on affective and behavioral engagement. Regarding the internal influencing mechanism, cognitive engagement positively drives affective and behavioral engagement, and affective engagement further exerts a positive influence on behavioral engagement. [Innovation/Value] Grounded in the research findings, this paper proposes three PDCS optimization strategies: adopting a user-engagement orientation, advancing culture-technology integration, and fostering a sustainable trust ecosystem.

[Purpose/Significance] As artificial intelligence increasingly integrates with intelligence analysis, traditional intelligence operations face transformational challenges toward intelligentization. How to systematically construct intelligence-oriented analytics systems has become a critical issue that urgently needs to be addressed. By deconstructing Palantir’s technical architecture and innovative practices, this study reveals the construction principles of modern intelligent analytics systems, providing valuable insights for the intelligent transformation of China’s intelligence agencies. [Method/Process] Combining survey interviews with literature analysis, this research employs a three-tiered analytical framework of “platform deconstruction → distinctive feature exploration → strategic recommendations," this research conducts an in-depth analysis of the core technical architecture, key capability systems, and implementation pathways of intelligence-oriented analytics systems. [Result/Conclusion] Through systematic integration of core technical elements including platform ecosystems, ontological reconstruction, microservice architecture, human-machine collaboration, and decision-driven approaches, Palantir has developed distinctive technical philosophies encompassing intensive architecture, cross-domain integration, modular assembly, progressive evolution, and goal-oriented analysis, achieving effective transformation from traditional data processing to intelligent cognitive analytics. This study proposes a construction framework for intelligence-oriented analytics systems, encompassing strategic pathways including scalable technical architecture, multi-source resource integration, modular capability systems, user-centric service systems, and decision-oriented analytical systems, providing a theoretical foundation foundation and practical guidance for advancing the modernization transformation of China’s intelligent analytics capabilities.

[Purpose/Significance] In order to adapt to the changes in university research and education among others in the era of digital intelligence, respond to the national initiative to enhance the digital literacy of the entire population and the IFLA’s call for improving librarians’ capabilities to address the current AI environment, strengthen academic librarians’ adaptability, competence, and creativity in the digital and intelligent context, and improve AI literacy across all business processes. [Method/Process] By reviewing relevant literature, AI literacy frameworks domestically and internationally, as well as the professional characteristics of university librarians, this study summarizes the main drivers compelling academic librarians to enhance their AI literacy, explores the core elements of AI literacy for academic librarians, and proposes relevant improvement paths based on these drivers and core elements. [Result/Conclusion] The changes in the AI-driven knowledge and information environment, library practices, and user needs are identified as the main drivers for enhancing academic librarians’ AI literacy in the era of digital intelligence. Based on this, the core elements of AI literacy for academic librarians are proposed, focusing on professional engagement, library management, library services, as well as teaching and guidance. Accordingly, a three-tier parallel AI literacy cultivation model—“foundation-core-extension”—along with related safeguarding measures are put forward.

[Purpose/Significance] To address limited intelligence, passive service models, and inefficient resource integration in library reference services, this study explores intelligent transformation pathways enabled by the Model Context Protocol (MCP). The lack of standardized interaction between large language models (LLMs) and heterogeneous library systems restricts the effective use of resources and professional services. By connecting LLMs with external resources and authorized tools through a unified protocol, MCP can help establish proactive, precise, personalized, and context-aware services and support smart library development. [Method/Process] Considering MCP’s characteristics and compatibility with library operations, this study constructs a four-layer interaction model of “Resource-MCP-LLM-User” and proposes an architecture comprising the MCP server, MCP protocol, foundational LLM, and user layers. The MCP server layer encapsulates library databases, knowledge bases, and tools as standardized resources and callable functions. The protocol layer supports capability discovery, permission control, and data exchange. The foundational LLM layer performs intent recognition, task planning, tool selection, information synthesis, and response generation, while the user layer supports natural-language interaction. Through protocol parsing, tool encapsulation, context management, and task coordination, the architecture enables LLMs to invoke heterogeneous resources and integrate information into coherent knowledge services. Service rules, domain knowledge, and access permissions can be incorporated to preserve librarians’ professional role. [Result/Conclusion] The findings indicate that MCP can improve the connectivity, interoperability, flexibility, and scalability of intelligent reference services. Through standardized and controlled interfaces, libraries can expose existing resources and capabilities to LLMs and expand services according to their collections, infrastructure, expertise, and user needs. MCP therefore supports a transition from passive question answering to proactive need identification, contextual interaction, and knowledge-based assistance. The proposed model provides a scalable theoretical framework and feasible technical guidance for library transformation, promotes the integration of resources, intelligent technologies, and professional services, and advances public cultural services toward proactive, personalized, and knowledge-driven development.

[Purpose/Significance] Open-source generative AI technologies, represented by DeepSeek, are accelerating the process of technological inclusiveness. However, user heterogeneity has led to significant differentiation in technology adoption paths. Therefore, it is urgent to abandon the homogeneity assumption of users in traditional models. Taking user heterogeneity as the core theoretical perspective and innovation, this study systematically deconstructs the technology adoption driving mechanisms for developers, enterprise users, ordinary users, and exploratory users. [Method/Process] The Technology Acceptance Model (TAM), the Unified Theory of Acceptance and Use of Technology (UTAUT), and the Information Ecology Theory are integrated. Based on users’ differences in the three dimensions of “role-capability-demand”, a “scenario-technology-ecology” analytical framework for open-source generative AI is constructed. A total of 391 valid questionnaire data were collected. Structural Equation Modeling (SEM) was employed to analyze the causal relationships between indicators, and fuzzy-set Qualitative Comparative Analysis (fsQCA) was utilized to explore the configurational path differences across the four user groups. [Result/Conclusion] The findings indicate that developers’ technology adoption intention is driven by the trade-off between technical efficiency and privacy, presenting two paths: “full-dimensional drive” and “risk avoidance”. Enterprise users’ acceptance is highly scenario-specific, pursuing four types of needs including “community empowerment” and “compliance priority”. Ordinary users rely on personal experience and pursue lightweight experiences. Exploratory users focus on lightweight experiences and novelty. The results show that user heterogeneity significantly influences technology adoption paths, and differentiated governance strategies are required to advance the inclusive application of open-source generative AI.

[Purpose/Significance] With the rapid proliferation of AIGC technologies, disparities in users' comprehension and application abilities have become increasingly pronounced, exacerbating the digital divide. This not only undermines the user experience but also affects trust in and acceptance of AIGC technologies. Investigating the relationship between users' algorithmic literacy and the perceived credibility of AIGC is crucial for optimizing human-AI interaction experiences and reshaping trust mechanisms. [Method/Process] Based on the Heuristic-Systematic Model and the Computers as Social Actors(CASA) theory, this study incorporates social attributes such as anthropomorphic expression and empathetic perception to move beyond traditional unidimensional design approaches. It adopts an algorithmic literacy framework covering five dimensions: algorithmic knowledge, awareness, evaluation, coping, and creation. It develops a theoretical model to examine its impact on the perceived credibility of AIGC, and employs structural equation modeling for empirical analysis. [Result/Conclusion] The findings indicate that algorithmic literacy enhances users' perceived information quality, thereby promoting systematic information processing and significantly increasing perceived credibility. Conversely, it suppresses heuristic processing by diminishing perceived functional quality, consequently exerting a negative impact on perceived credibility. User characteristics such as gender and educational background do not significantly moderate these paths. Algorithmic knowledge, evaluation, and creation positively influence perceived information quality; algorithmic awareness and knowledge negatively affect perceived functional quality; while algorithmic knowledge and coping positively influence perceived AI ethics.

[Purpose/Significance] This study investigates the effect of digital reverse mentoring on online health information acquisition among older adults and its underlying mechanisms. It aims to offer insights into addressing digital poverty and enhancing older adults' digital health literacy. [Method/Process] An information-seeking task was designed for a quasi-experiment involving 42 older adults. The task completion outcomes of the experimental group were compared with those of the control group. The mechanism by which digital reverse mentoring influences online health information acquisition among older adults was examined through dual pathways employing grounded theory coding analysis based on the interview data. [Result/Conclusion] For simple tasks, digital reverse mentoring enhances the efficiency and quality of health information acquisition among older adults. For complex tasks, its effect is more pronounced, significantly improving the quality and accuracy of health information acquisition. However, efficiency decreases due to the time-intensive nature of the process, characterized by “slow decision-making, high quality”. Through the dual pathways of “individual intervention” and “resource support”, digital reverse mentoring influences older adults by reconstructing their cognitive and literacy abilities, improving their health decision-making, and consequently enhancing their information acquisition effectiveness.

[Purpose/Significance] Null models are crucial tools in network analysis for controlling potential confounding factors and reducing bias in research conclusions. This study aims to propose efficient and scalable randomization strategies to improve the feasibility and accuracy of null models when applied to large-scale network analyses. [Method/Process] We systematically review the foundational principles and methodologies of null models, discussing their advantages and limitations and identifying appropriate research scenarios for their application. Building upon the Fisher-Yates reshuffle algorithm, we develop a suite of efficient randomization algorithms tailored to common constraints in citation networks, such as node attributes, preferential attachment, and the exclusion of self-loops. The reshuffle strategies are further optimized by integrating a random value optimization method and a max-heap data structure, which ensures strict adherence to constraints while reducing time complexity to O(Mlog M). Furthermore, through an empirical analysis of citation preferences among journals in the library and information science field, we compare outcomes generated by different null models to demonstrate the effectiveness of the proposed approaches. [Result/Conclusion] Our findings indicate that the selection of an appropriate null model must correspond to the specific research question. This work indicates that the choice of a null model must align with the research question(s): simple models may lead to “straw man” conclusions, while overly complex models risk overfitting. Moreover, different null models can produce varying results, emphasizing the necessity for cautious model selection grounded in the research context. This work discusses three scenarios of constraints; future research directions include the examination of specialized constraints and their combinations to broaden applicability.

[Purpose/Significance] This study conducts quantitative identification and impact evaluation of sleeping beauty papers from the perspective of full-text bibliometrics, provides new identification indicators and methods for the quantitative identification of sleeping beauty papers, and helps high-value sleeping beauty papers play a role in promoting scientific development. [Method/Process] This study utilizes papers published in the fields of management and computer science between 1996 and 2014 in the Web of Science database as the data source. First, an improved YCPA index is employed to quantitatively identify sleeping beauty papers, with the identification effectiveness being validated. Second, from the perspective of full-text bibliometrics, an evaluation index system for the impact of sleeping beauty papers is constructed, encompassing three dimensions: academic impact, social impact, and perceived impact. Third, the entropy weight method is applied to conduct empirical single-dimensional and comprehensive evaluations, and the evaluation results are verified using the BP neural network model. Finally, by analyzing the evaluation results of disciplinary differences in the characteristics of sleeping beauty papers, the features underlying the sleeping and awakening of them are examined. [Result/Conclusion] The conclusions are as follows.(1) The improved YCPA index has a good recognition effect on sleeping beauty papers.(2) The influence evaluation model of sleeping beauty papers constructed from the perspective of full-text bibliometrics has achieved effective signal and comprehensive evaluation. The social impact has the strongest positive influence on the comprehensive impact of sleeping beauty papers. Among various indicators, team strength, interdisciplinary span, and journal influence significantly enhance the academic impact of sleeping beauty papers. Official policies and public discussions on social media platforms significantly increase social impact. The mention of emotions in the cited texts significantly enhances the perceived impact.(3) This study analyzes the features of sleeping and awakening of sleeping beauty papers in management and computer science.

[Purpose/Significance] To uncover the driving mechanisms of individual conversational health information search behavior by examining the roles of health literacy (higher vs. lower), AI trust (higher vs. lower), and task type (acute illness vs. chronic illness). The findings provide a theoretical basis and practical implications for the design of human-AI collaborative health service decision-making systems. [Method/Process] A laboratory experiment was conducted with 72 participants performing health information search tasks. Mann-Whitney U tests and Scheirer-Ray-Hare tests were employed to analyze differences in search behavior indicators and to propose the underlying driving mechanisms. [Result/Conclusion] Significant main effects and interaction effects of health literacy, AI trust, and task type were found across multiple conversational health information search behavior indicators such as “average query length” and “dialogue rounds”. Individuals with both high health literacy and high AI trust demonstrated the longest average query length, while those with low levels of both showed the shortest. Participants with higher AI trust engaged in significantly more dialogue rounds during acute illness tasks than in chronic illness tasks. In contrast, for individuals with lower AI trust, the “follow-up explanation-seeking behavior”, “dialogue rounds”, and “question relevance” in chronic illness tasks were significantly higher than those in acute illness tasks.