Purpose/significance Addressing prevalent misconceptions and ambiguous understandings between intelligence and think tanks in current academic discourse，this paper，from the perspective of front-line intelligence workers，elucidates and differentiates these concepts while clarifying their identities，aiming to provide references for domestic research and practice development. Method/process Through tripartite analysis encompassing definitions，functions，and future focal points for work，this study interprets and distinguishes the “convergence”“transformation” and “differentiation” between intelligence and think tanks，thereby clarifying their essential connotations，operational effectiveness，core operational priorities，and historical missions. Result/conclusion The evolution of intelligence think tankization and think tank informatization does not constitute revolutionary innovation nor the pursuit of entirely novel operational paradigms. Rather，it emphasizes elevating previously neglected or underdeveloped functionalities to more prominent positions. Instead，they aim to highlight certain functions that have been overlooked or weakened，so that users and decision-makers can deeply associate entities with their inherent connotations and effectiveness. ［ Originality/value ］From the perspective of frontline intelligence practitioners and within the context of “transformation” and “integration” as dual core tasks，this paper provides theoretical foundations and practical guidelines for the future “ization” of intelligence-think tank integration and think tank informatization through definition deconstruction，efficacy comparison，and analysis of focal implementation points.

Purpose/significance To address monitoring deficiencies in intelligence-cognitive environments during cognitive gaming scenarios，this study engineers a scanning-monitoring genealogy framework for intelligence-cognitive ecosystems. Method/process Through case-based analysis and theoretical synthesis，this research examines three constitutive elements：information sources，transmission vectors，and content architectures. Building upon this foundation，we propose the Intelligence-Cognitive Environment Scanning-Monitoring-Response（ICESMR） model. Result/conclusion Integrating cognitive warfare theory and intelligence fact theory，we delineate a four-tier monitoring genealogy spanning from facts through leads and signals to weak signals. The ope rationalized ICESMR model enables risk signal identification and triggers mission-adaptive responses，ultimately enhancing situational threat mitigation.

Purpose/significance This research aims to investigate the impacts，characteristics，and manifestations of biased narratives in foreign media，as well as the intrinsic mechanism through which such narratives exert influences on China’s intelligence assessment. Its core objective is to uphold the objectivity and neutrality of China’s intelligence assessment process. Method/process Via literature synthesis and case analysis，this study identifies biased narrative strategies of some foreign media. Integrating Framing Theory and Reflexive Control Theory，it delineates the mechanism of biased narratives and proposes targeted mitigation strategies. Result/conclusion Macroscopically，biased narratives follow a three-stage process：issue securitization，role formulation，and material organization. They contaminate the information ecosystem，distort intelligence assessment’s cognitive framework，disrupt its verification logic，and reinforce/mislead intelligence cognition. In intelligence assessment，deconstructing biased narratives and reconstructing intelligence facts is imperative to counter their adverse impacts.

Purpose/significance Analyze the elements，matters，and formation mechanisms of cognitive biases in the scenarios of cognitive game in intelligence field，propose countermeasures against cognitive biases，enhance the initiative in cognitive game，and strengthen the cognitive resilience of the intelligence system. Method/process Based on the understanding of intelligence facts，sort out the typical manifestations of cognitive biases in the scenarios of intelligence cognitive gaming，systematically analyze the formation mechanisms of cognitive biases，and propose countermeasures. Result/conclusion The formation and evolution of cognitive biases in intelligence involve self-sustaining feedback and amplification loops through three stages：cognitive environmental pollution，individual cognitive anchoring，and group cognitive shaping，which enables the implantation，formation，and solidification of cognitive biases. Therefore，cognitive enhancement strategies are proposed from two aspects，namely，intelligence individuals and the intelligence system，to address the phenomenon of cognitive biases in intelligence game.

Purpose/significance AI governance is a necessary guarantee for the healthy development of artificial intelligence. Current research on AI governance，both domestically and internationally，faces two major limitations：insufficient systematic analysis of risks and structural deficiencies in governance path，urgently necessitating the construction of governance path adapted to the development of AI technology. Method/process By selecting literature abstracts from WoS and CNKI，this study used the BERTopic model to identify risk types of common concern in domestic and international research，analyzing risk interaction relationships，and constructing a governance path. Result/conclusion The objects of AI governance include two types of risks： one type consists of endogenous risks caused by technical defects occurring during data processing and model training phases；the other type comprises risks derived from endogenous risks，including abuse，infringement，difficulty in accountability，cognitive harm，and ethical anomie. Based on the analysis of the interaction between endogenous and derived risks，as well as the dereliction of duty issues and deficiencies in collaborative mechanisms among multiple participants throughout the AI lifecycle，we integrate the concepts of “human-in-the-loop”，multi-stakeholder collaborative governance，and agile governance to establish a closed-loop governance path of “legal traction-technical regulation-dynamic auditing-accountability”，providing a reference for the development of a Chinese-style AI governance system.

Purpose/significance Developing a classification framework for the application value of data elements helps clarify the value they generate in different scenarios，thereby providing theoretical support and practical guidance for identifying and unlocking their potential. Method/process Adopting a mixed research paradigm that integrates “theory-driven” and “data-driven” approaches，this study draws on the information technology value theory and employs literature review，open-ended questionnaires，and the Delphi method to develop an initial framework from four dimensions：individual，organizational，inter-organizational，and societal. Subsequently，open and closed card-sorting methods are used to validate the structural and logical relationships among categories，and further refine the framework. Result/conclusion The final framework includes 4 first-level categories，22 second-level categories，and 70 third-level categories. Card-sorting results indicate good discriminability among categories，confirming the framework’s scientific validity. By focusing on “value release”，this study not only enriches the current research on the value of data elements，but also provides a theoretical foundation and practical guidance for unlocking the potential of data elements.

Purpose/significance With the increasing prevalence of data-intensive research paradigms，analyzing the factors influencing the reuse of scientific data and exploring effective pathways to promote such reuse is of paramount importance. Understanding these dynamics is crucial for unlocking the value of scientific data，driving scientific innovation，and enhancing economic benefits. Method/process Based on a theoretical model of the factors influencing scientific data reuse，this paper develops a configuration analysis framework for identifying the driving pathways of scientific data reuse. Utilizing 2858 datasets from 20 data repositories，we conduct an in-depth analysis of the multiple concurrent causal mechanisms that promote scientific data reuse. This analysis employs necessity analysis and fuzzy-set qualitative comparative analysis （fsQCA） methods. Result/conclusion The study reveals that scientific data reuse results from the synergistic effect of multiple factors. None of the six conditional variables within the four dimensions of data accessibility，quality，context，and trust are necessary conditions for high scientific data reuse. We identify five distinct configuration pathways that promote scientific data reuse，which can be categorized into three models：“Data Quality-Trust Driven”“Data Quality-Context Driven” and “Data Accessibility-Quality-Trust Synergistic”. The diversity of data tags and associated literature play crucial roles in facilitating scientific data reuse.

Purpose/significance Against the backdrop of diversified international competition and complex intelligence games，the necessity and urgency of counterintelligence have become increasingly prominent. A systematic review of counterintelligence research progress is conducive to guiding future related research and practice. Method/process Based on the big intelligence view，this study selected 301 domestic and international journal articles related to counterintelligence using the PRISMA method as research objects. It conducted a descriptive analysis of research status，systematic analysis of conceptual connotation，theories and models，application and implementation of counterintelligence，as well as existing challenges. Result/conclusion The study clarifies the conceptual connotation and research boundaries of counterintelligence，alleviates cognitive biases caused by temporal-spatial differences to some extent，and provides a consistent theoretical foundation for understanding and applying counterintelligence in new environments and contexts. Limitations This study lacks adequate attention to relevant findings presented in conferences，books，and other academic publications，which should be addressed and supplemented in future studies.

Purpose/significance Against the backdrop of the deepening evolution of great power rivalry，based on the capabilities and security needs of China’s counterintelligence work，this paper specifically analyzes the value and basic path of DeepSeek empowering China’s counterintelligence work，providing theoretical support and practical reference for building an autonomous and controllable counterintelligence technology ecosystem. Method/process Based on the perspective of the triadic interaction of technology，power，and security，this paper specifically analyzes the cumulative impact of great power rivalry and technological innovation on the intelligence ecosystem，the technological breakthroughs of DeepSeek and counterintelligence value，as well as the operational models，risks and challenges of DeepSeek in empowering China’s counterintelligence work. Result/conclusion Faced with the increasingly intense great power rivalry，China’s counterintelligence work must gear up for the intelligent era. This entails leveraging domestic large language models，represented by DeepSeek，to empower operational systems. This involves exploring the construction of a counterintelligence work platform that integrates adversary intelligence collection，risk early warning，decision support，and attack and defense handling. At the same time，it is necessary to continuously research and solve the risks and challenges faced by the application of artificial intelligence technology in the field of counterintelligence，and enhance the security of the system.

Purpose/significance This paper explores the impact of AIGC （AI Generated Content） technology on the transformation of library governance paradigms，and promotes the intelligent transformation and upgrading of library governance systems. Method/process This paper adopts methods such as theoretical analysis，literature research，and comparative study，integrates multi-subject collaborative governance theory，technology ethics theory and library science theory，and breaks through the limitations of the traditional “technology-governance” dual structure. Study on the impact of AIGC technology on library boundary breakthrough，service model optimization，information retrieval ecosystem reconstruction，and information management theory innovation. By constructing a “technology-institution-governance” ternary collaborative governance model and a dynamic response mechanism，this paper proposes a multi-dimensional interactive driving logic of “technology iteration-risk transmission-institution response-governance upgrading”，and forms a positive cycle where technology iteration drives institutional optimization，institutional improvement promotes governance upgrading，and governance enhancement feeds back to technological innovation. Result/conclusion AIGC （AI Generated Content） technology has brought new opportunities for the intelligent transformation of library governance. It resolves the contradiction between efficiency and security through technological innovation，balances the relationship between innovation and ethics via institutional optimization，and enhances dynamic response capabilities through governance reform. This provides practical paths and theoretical references for libraries to address the risks brought by AIGC technology and improve governance efficiency.

Purpose/significance With the rise of generative large language models （LLMs） applications，users' information acquisition channels are gradually shifting toward intelligent Q&A platforms. This transformation generates “multi-turn dialogue” sequential data，which can serve as objective evidence for evaluating user search effectiveness. Method/process Based on a research context，this study explores the underlying principles of information utility in user information-seeking behavior. Integrating the Expectation-Disconfirmation Theory （EDT），we propose an information utility measurement model and conduct controlled experiments to assess the value of information utility across users with varying cognitive backgrounds. This study employs a combination of experimental research and quantitative data analysis to dissect multi-turn interactions between users and LLMs. Using the Accumulate-Utility metric，we examine how the gap between expectations and actual experiences influences utility fluctuations during information-seeking processes，thereby validating the proposed measurement mechanism. Result/conclusion The information utility framework effectively explains user preferences，experiential outcomes，and the resulting utility levels derived from their discrepancies. It also reveals variations in information utility across tasks with different attributes，further refining the methodology for analyzing information-seeking effectiveness.

Purpose/significance Health misinformation on social media is pervasive and detrimental，necessitating a clearer understanding of how users perceive and respond to such information. This study focuses on users’ cognitive processing mechanisms to enhance understanding of individual health misinformation processing and provide a theoretical foundation for designing targeted intervention strategies. Method/process Drawing on the critical incident technique and scenario-based experimental simulations，interview data were collected from 25 users and analyzed using the Gioia methodology to conduct coding analysis and model construction. Result/conclusion Social media users process health misinformation through two primary cognitive pathways： intuitive health heuristics and analytical health deliberation. The former includes perceptions of information sources，social endorsement，narrative style，personal preferences，and algorithmic beliefs. The latter involves information verification，complex reasoning，cognitive involvement，and experiential processing. The interaction of two pathways gives rise to three distinct cognitive judgment states—disbelief，uncertainty，and belief—which trigger information behaviors，including fact-checking，information seeking，adoption，and sharing. ［Innovation/ value This study deepens the understanding of user interactions and decision-making processes with health misinformation，enriches the specific processing ways under different cognitive paths，and provides a theoretical reference for subsequent targeted information governance.

Purpose/significance By identifying frequent local interaction patterns in patent networks，defining and calculating the importance of patents in higher-order network structures，enriching patent network analysis methods，and providing a new perspective for understanding the structure and evolution of patent networks. Method/process Firstly，define and extract the feedforward loop phantom structure in the patent network. Secondly，the higher-order structural features are linked with the importance of patents，and based on this，two patent evaluation models PCDRank and PCSRank based on feedforward loop phantom are proposed to identify the core patents. Finally，the information dissemination dynamics model is used to calculate the influence of patent dissemination，and the recognition effect of the proposed model and other centrality indexes is verified. Result/conclusion The higher-order structure of patent citation network contains richer structural information，which is helpful to increase the amount of patent analysis information. The core patent recognition method based on high-order network structure can obtain more accurate recognition results. The dynamics model of information propagation is used to verify the validity and robustness of the proposed identification model. ［Innovation/ value By combining high-order structural characteristics with global network topology characteristics，a more accurate core patent identification framework is constructed，and a new perspective is provided for understanding the structure and evolution of patent networks.

Purpose/significance Sorting out the evolution mechanism and key motivations from the perspective of public opinion reversal，and putting forward efficient measures to deal with public opinion reversal. Method/process Firstly，the index system is constructed. Based on the grounded theory，the case data are collected to build an index system of influencing factors for the reversal of online public opinion of hot events. With the help of system dynamics method，this paper constructs an evolution model of online public opinion reversal in emergencies，and analyzes the influence of different types of variables on the heat of public opinion reversal. Finally，simulation analysis. The event of “21-year-old boy ‘Fat Cat’ jumping into the river and committing suicide” is selected to observe the influence trend of different values of each variable on the evolutionary system，and reveal the key motivation and evolution mechanism of the reversal heat of online public opinion of hot events. Result/conclusion The evolution process of online public opinion reversal in hot events is divided into four stages： initiation，fluctuation，reversal and calm down. From the perspective of the system as a whole，public opinion events and government control have a high impact on heat. Accordingly，combined with the simulation results，countermeasures are put forward to provide theoretical reference for reducing the harm of online public opinion reversal.

Purpose/significance While Generative Artificial Intelligence technology reshapes information production models，it also gives rise to practical issues such as the proliferation of disinformation and disorder in cyberspace. Implementing labeling for AIGC has thus become a global regulatory consensus. Although China has initially established a regulatory framework for AIGC labeling，the existing rules still fail to eliminate the phenomenon of “label fatigue”. Method/process By adopting literature analysis and case study methods，and within the framework of the Theory of Cognitive Load in Information Processing and information architecture，this study explains the connotation and causes of AIGC “label fatigue”，and explores paths for improvement from the perspective of hierarchy and classification. Result/conclusion AIGC “label fatigue” stems from four factors：the lack of standardization in labeling，the fragmentation of labeling standards，the mismatch between label design and users’ cognitive abilities，and the disconnection between label functions and the needs of specific scenarios. In response，it is necessary to adhere to five fundamental principles：user-centricity，necessity and appropriateness，clarity and readability，consistency and standardization，and dynamism and adaptability. Furthermore，starting from four dimensions—AIGC hierarchical standards，classification systems，specifications for labeling elements，and presentation strategies—this study proposes constructing a set of hierarchical and classified labeling rules for AIGC that centers on users’ information cognition and is oriented toward the accurate transmission of risks.

Purpose/significance This study proposes a method for constructing a dynamic fusion network DAG of “science–technology” topics by integrating BERTopic-based topic semantics and transition probabilities，and employs dynamic programming to extract domain-specific topic evolution paths. It offers a systematic reference for optimizing the layout of industrial technology research and development. Method/process Through knowledge-augmented BERTopic temporal topic modeling，a dynamic topic fusion network is constructed based on semantic similarity，transition probability，and the importance of topic nodes. Graph traversal algorithms are applied to extract the evolution path of “science–technology” topics. Result/conclusion An empirical study in the field of lithium extraction from salt lakes demonstrates that the proposed method can accurately identify the evolution path of “science–technology” topics within the domain，revealing the dynamic development process from basic research to technological application and transformation. This provides valuable reference for understanding the advancement of science and technology in the field. ［Innovation/ limitation The study innovatively combines BERTopic temporal topic modeling，complex networks，and graph traversal techniques to extract domain-specific topic evolution paths. Future research could incorporate multi-source data such as policies and market information，along with citation features，to achieve a more holistic analysis of the entire life-cycle evolution of science and technology.

Purpose/significance To integrate cognitive neuroscience tools into the user experience map，optimize the theoretical framework of user experience evaluation of university mobile library，and provide new ideas and methods for the study of user experience of university mobile library. Method/process Using eye tracking，electro skin monitoring，electro brain monitoring and other cognitive neuroscience tools to optimize the construction process of traditional user experience map，then designing experiments which taking the mobile borrowing service of a “Double First Class” university mobile library as an example to accurately analyze the online borrowing service experience of university mobile library users，also validating the effectiveness of the user experience map after process optimization. Result/conclusion The results show that the cognitive load and emotional reactions of university mobile library users in mobile borrowing services change symmetrically，which is consistent with the user interview feedback. The theoretical framework of user experience map integrating cognitive neuroscience tools is verified to be effective in measuring emotional changes and real needs in the process of user experience. The results provide scientific basis and tool guidance for future user experience evaluation research and service path optimization practice in complex scenarios.

Purpose/significance The identification of weak signals of disruptive technologies is crucial for reshaping industrial landscapes and driving the development of new productive forces，and it is a key path to solving the problem of high uncertainty in technological innovation. Method/process A method for predicting disruptive technologies that integrates generative language models and multi-dimensional features is proposed. Firstly，a multi-dimensional weak signal indicator system is constructed，and prompt fine-tuning of generative large language models is used to parse patent abstracts and extract semantic features of technical contributions. Secondly，a Stacking ensemble learning framework is built to optimize classification boundaries，and the SHAP model is used to quantify feature contributions. Result/conclusion Empirical evidence shows that this method has significant predictive performance in the fields of fuel cells，artificial intelligence，and quantum computing. Among them，the degree of strategic planning and technological breakthroughs has the highest contribution importance，and interdisciplinary nature presents a "double threshold effect". This finding verifies the effectiveness and rationality of the model and provides an integration paradigm of semantic parsing and structured indicators for early technology signal capture. ［Innovation/ value This method provides a basis for the precise allocation of innovative resources，and the new paradigm can be extended to multiple fields，offering implementable methodological support for addressing the uncertainty in technological innovation.

Purpose/significance This study aims to overcome the technical bottlenecks of fine-grained sentiment analysis in multimodal online public opinion during sudden natural disasters，breaking through the semantic limitations of unimodal text analysis. It seeks to provide management authorities with a theoretical toolkit that integrates both risk perception capture and dynamic guidance decision-making to address the phenomenon of public opinion entropy increase. Method/process This study constructs a cross-modal sentiment recognition model integrating RoBERTa and Vision Transformer（ViT），and introduces BERTopic-based topic classification to achieve dual-granularity “topic-sentiment”analysis. The effectiveness of the proposed model is validated through comparative experiments on empirical data. Result/conclusion The cross-modal dynamic fine-grained sentiment analysis method proposed in this study significantly enhances the accuracy of public opinion sentiment analysis，while simultaneously revealing the differentiation patterns of group cognition under topic-sentiment coupling effects. The proposed RoBERTa-ViT model achieves an F1-score of 0.8043，outperforming conventional unimodal models by 15%～24%. It successfully identifies six major themes（e.g.，disaster alerts，institutional reflection）and their corresponding sentiment distributions. ［Innovation/ limitation This study is rooted in the urgent need for the governance of fragmented and heterogeneous data in social media scenarios. It breaks through the traditional public opinion analysis path that mainly focuses on single-modal static analysis，and constructs a RoBERTa-ViT multimodal sentiment recognition framework. Through empirical research，it verifies the traction effect of thematic dimensions on emotional polarization.

Purpose/significance In response to the cognitive disruption and decision-making risks caused by the spread of misinformation，this study explores a structured practical approach to intelligence verification，aiming to enhance the reliability and explainability of the verification process in real-world scenarios and provide methodological support for intelligence professionals in addressing misinformation challenges. Method/process In this paper，a descriptive intelligence verification framework based on evidence chains and large language models is proposed，encompassing five key components： claim detection，evidence retrieval，evidence chain construction，claim verification，and explanation generation. Customized prompt engineering is employed to guide large language models in producing the required structured outputs，enabling stable and precise task control. Result/conclusion The empirical results in the field of health informatics demonstrate that the proposed framework can effectively integrate multi-source evidence，accurately verify the authenticity of claims，and generate high-quality explanatory texts，thereby providing a viable technical pathway for descriptive intelligence verification in complex contexts. This study proposes the structured framework for descriptive intelligence verification，addressing gaps in procedural systems and chain-based reasoning，and improving the accuracy and trustworthiness of large language model–based verification in real-world scenarios.

Purpose/significance Under the background of the rapid development of artificial intelligence technology，exploring the representation，extraction，organization and application of scientific knowledge enabled by large language models is of great significance for the innovation of related theoretical methods and applications such as intelligent perception of scientific and technological information and decision support. Method/process Firstly，the PDMR-ER representation model of scientific knowledge is proposed，and then the entities，relationships and attributes of the PDMR-ER representation model of scientific knowledge are extracted and calculated by using the DeepSeek-R1-0528 through the Prompt project. Finally，the knowledge map is constructed by using the Neo4j graph database，and on this basis，the application scenario is explored by taking the prediction of scientific knowledge network links as an example. Result/conclusion Empirical evidence shows that large language models can effectively promote the deep semantic mining and organization of scientific knowledge. Combined with the Neo4j knowledge graph database，they can efficiently and accurately reveal potential links and innovation points of scientific knowledge，enabling AI-driven scientific knowledge services.

Purpose/significance This study compares the similarities and differences between AI-generated responses and government officials’ responses in government-citizen interaction scenarios from three dimensions：semantic similarity，language patterns，and thematic features. It aims to provide decision support for optimizing intelligent government-citizen interaction. Method/process A total of 1646 pieces of government-citizen interaction data were collected from the official website of Z Provincial Government. Citizen messages were input into the DeepSeek-R1 model to generate response texts. For both government officials’ responses and DeepSeek-generated responses，the TF-IDF and Doc2Vec models were applied to compare their word-level and document-level semantic similarity；the DeepSeek-R1 model was used to classify the language patterns of the two types of response texts；and the BERTopic model was employed to analyze the relationships between topics，between topics and topic keywords，and between topics and documents. Result/conclusion The word-level and document-level similarity between DeepSeek responses and government officials’ responses are both low，at 0.03 and 0.33，respectively. However，the two types of responses show little difference in word selection and semantic expression when addressing different questions. Both comprehensively utilize the three language patterns，but the proportion of each pattern in DeepSeek responses is unbalanced：empathetic discourse accounts for only 16.8%，indicating a relatively weak emotional connection with the public. Compared with government officials’ responses，DeepSeek responses have more dispersed themes，fewer thematic layers，higher semantic representation effectiveness of thematic keywords and greater semantic deviation—suggesting room for improvement in the stability of response quality. Additionally，across 18 government-citizen interaction domains，DeepSeek responses exhibit uneven consistency in themes and documents. Based on these findings，insights are provided to enhance the efficiency and public trust of intelligent government-citizen interaction services.