Purpose/significance As the external constraints governing economic activities shift from an efficiency-oriented market equilibrium to security-oriented political contestation，national economic security is no longer confined to the cultivation of developmental advantages and the restoration of economic order. It increasingly involves the mitigation of external vulnerabilities and the safeguarding of critical channels. Against this background，relevant knowledge production should be grounded in a holistic approach to national security，particularly the principle of coordinating development and security. To address the complex challenges facing national economic security governance，scholarly inquiry must move beyond single-disciplinary analysis toward interdisciplinary integration. Method/process This article conceptualizes “national economic security intelligence” as research object，and elaborates the interdisciplinary intersection of economic research and intelligence research in the core academic propositions and foundational methodological systems. Result/conclusion For core academic propositions，the two fields intersect around risk early warning，situational awareness，and policy evaluation. For foundational methodologies，this article adopts construct validity，internal validity，and external validity as its analytical framework，and examines the complementary logic of the two fields in the measurement of object attributes，the identification of causal relationships，and the evaluation of conclusion validity. Taking the construction of international trade resilience amid escalating export controls as an illustrative scenario，this article further shows how such interdisciplinary integration can guide concrete governance practice，thereby providing an indigenous economic knowledge framework for the deeper engagement of Chinese intelligence studies in national security governance.

Purpose/significance The dormancy of scientific research big data is a process that cannot be ignored in the development of scientific research big data. Measuring the degree of dormancy of scientific research big data helps to better understand the current status of dormancy of scientific research big data from different aspects and provides a certain reference for the subsequent value recovery and reuse of dormant scientific research data. Method/process This paper presents the concept and benchmark formula of the dormancy degree of scientific research big data. On this basis，it analyzes and calculates the journal dormancy degree/disciplinary dormancy degree，complete/incomplete dormancy degree，and annualized average dormancy degree. It selects literature included in CNKI from 2005 to 2014 as samples for empirical analysis，and based on this，explores the potential laws contained in the dormancy of scientific research big data. Result/conclusion The results show that the dormancy of scientific research big data exists in different disciplinary fields，different authors，and different journals，with varying depths of dormancy. A systematic and multi-dimensional understanding of the dormancy degree of scientific research big data is conducive to maximizing the excavation of the deep value of dormant data and promoting the sound development of the scientific research big data ecosystem.

Purpose/significance Data pollution enrichment effect is a harmful phenomenon that occurs in data activities. As the flow of polluted data in the “data chain” and the increase in data levels，the degree of enrichment continues to increase. Exploring its internal mechanism is of great significance and value for effective data governance. Method/process Based on the theory of enrichment effect in ecology，this article proposes the concept of “data pollution enrichment effect”，analyzes its characteristics，constructs the “overall model of data pollution enrichment effect” （OM-DPEE），deeply analyzes its internal mechanism，and combines practice to provide an evolutionary path for data pollution enrichment effect，thus providing assistance for data governance in related fields. Result/conclusion The study shows that the data pollution enrichment effect is a dynamic process in which the enrichment of harmful data increases with the flow of data in the "data chain" and the increase of data levels，causing harm to subsequent data activities. Its dimensions include intensity dimension，structure dimension，spatial dimension，time dimension，hierarchical dimension，etc. Its evolution path includes “weak enrichment→strong enrichment”“linear enrichment→nonlinear enrichment”“internal enrichment→external enrichment”“occasional enrichment→long-term enrichment”“single level enrichment→multi-level enrichment”“artificial enrichment→AI enrichment” and other types. And the above paths form intertwined and intricate relationships.

Purpose/significance When Artificial Intelligence，especially Generative Artificial Intelligence，becomes an indispensable tool for human life，work，and learning，intelligent outsourcing is becoming a common phenomenon of human intelligence interaction，and the relationship between humans and intelligent agents has become an urgent theoretical and practical problem that needs to be solved. Therefore，it is of great theoretical and practical significance to clarify the intersubjective structure of intelligent outsourcing under the context of GAI，identify its challenges，propose its resolution path，and understand and optimize the relationship of human intelligence interaction from the perspective of intersubjective theory. Method/process Based on a detailed review of related works on GAI empowerment，GAI subjectivity and objectivity，the inevitable transformation trend is revealed that the relationship between subject and object in intelligent outsourcing is shifting from the dichotomy between subject and object to their intersubjectivity. Based on the theory of intersubjectivity，the intersubjective structure of intelligent outsourcing is proposed and its operation mechanism is given in detail under the context of GAI，the intersubjective dilemma is presented in current human intelligence interaction，and the corresponding resolution path are proposed. Result/conclusion The intersubjective structure of intelligent outsourcing under the context of GAI includes three levels，which are interactive symbiosis，meaning co-construction，and responsibility sharing. The imbalance of interactive symbiosis，inaccurate meaning co-construction，and ineffective responsibility sharing are the implementation difficulties of intersubjectivity in intelligent outsourcing from theory to practice. It is necessary to explore the resolution strategies for intersubjective dilemma from the aspects of technology，ethics and system，especially from the inclusive characteristics of excellent Chinese culture to build a new pattern of harmonious and sustainable human intelligence relationship.

Purpose/significance Clarifying the micro-level mechanisms through which scientific knowledge flows into technology and examining how authority and novelty jointly reduce barriers to the technological translation of scientific findings are of great significance for optimizing the allocation of scientific and technological resources and improving the effectiveness of the national innovation system. Method/process Taking biochemistry and molecular biology as the research sample，this study constructs paired data consisting of real and virtual citation relationships to simulate the process by which patents select scientific papers for citation. The XGBoost algorithm is used to build the prediction model，and the SHAP interpretability framework is introduced to quantitatively analyze the complex mechanisms through which novelty and authority affect technological translation from three dimensions： main effects，interaction strength，and interaction patterns. Result/conclusion The results show that authority signals serve as a core driver of technological translation，with paper authority and journal authority jointly contributing more than 36.01% of the model explanation. Compared with its direct effect，the influence of novelty on technological translation is reflected more strongly in its interactions with other factors. The endorsement effect of authority is most robust in the medium-to-high novelty range，where it can effectively reduce screening and validation costs in technological contexts，but it may become less effective in the extremely high novelty range. ［Innovation/ value This study confirms that authority can，to some extent，increase the likelihood that novel scientific findings will be adopted in technological development，and provides empirical evidence for understanding how scientific novelty is translated into technological applications.

Purpose/significance As research activities increasingly shift toward computationally intensive and practice-oriented paradigms，source code，software tools，and open-source projects have emerged as important research outputs in fields such as computer science. However，traditional citation-based evaluation systems and existing altmetrics studies struggle to capture the actual impact of academic research in technological practice. To address this limitation，this study explores the potential of open-source code collaboration platforms as a new source of altmetric data. Method/process Taking GitHub as a case study，this study adopts a phenomenological analysis approach and constructs an analytical framework based on three dimensions： knowledge association patterns，association carriers，and citation forms. Based on this framework，citation relationships between code repositories and academic papers are identified and classified. Result/conclusion The results show that paper citations on GitHub exhibit multiple knowledge association patterns，including theory-driven implementation，practice-driven endorsement，collaborative innovation，and tool-oriented aggregation. These patterns reflect the diverse roles that academic knowledge plays in engineering practice. Open-source code collaboration platforms can help reveal the transformation pathway of academic research from theoretical proposals to technical implementation，providing a complementary perspective for extending altmetrics to technology-oriented impact assessment.

Purpose/significance This study identifies the key governance elements involved in the construction and operation of trusted data spaces from the perspective of participating actors，with the aim of providing theoretical support and practical guidance for advancing trusted data spaces from pilot exploration to normalized operation. Method/process Taking the typical construction practices of trusted data spaces in Jiangsu Province as the empirical case，this study adopts interviews and qualitative research methods to inductively analyze the relevant elements. On this basis，it constructs a governance framework for trusted data spaces composed of three levels—foundational conditions for construction，operational capabilities，and regulatory and standard mechanisms—and examines the internal relationships among them. It then further proposes corresponding optimization pathways. Result/conclusion The governance of trusted data spaces still needs to be advanced in three aspects：calibrating capability boundaries and strengthening scenario-driven design； selecting technologies in accordance with data characteristics and implementing tiered trusted protection；and refining rule systems to regulate the order of data circulation.

Purpose/significance The rise of the open-source LLMs deeply reshapes the risk ecosystem of cross-border data flows，rendering the traditional data governance frameworks based on physical perimeters and static compliance increasingly inadequate. Method/process Through the analytical perspective of data lineage，considering the risk events and industrial logics，this paper deconstructs the lifecycle of open-source large models into three stages： upstream （dataset construction），midstream （model training），and downstream （applications deployment）. It categorizes six major risk scenarios and proposes corresponding governance strategies. Result/conclusion The study finds that open-source large models face the risks of important/sensitive data export and data poisoning in the upstream； the risks associated with using computing power transnationally and outflow of model parameters in the midstream； and the risks of interaction prompt leakage and weaponization of derivative models in the downstream. Accordingly，the agile governance of open-source large models should focus on the transformation of data states，introducing new strategies as block chain-based metadata authentication，security detection models，privacy preserving computation，model lineage cards，semantic firewalls，and adjustment of open-source license，to enhance the granularity of data governance and empower the healthy development of technology enterprises in China.

Purpose/significance To explore the cognitive evolution mechanism of the public in the context of the rapid diffusion of AI technology，and reveal the dynamic formation process of AI literacy driven by technological booms. Method/process Taking the online discussions triggered by OpenClaw as the research context，this study selects relevant discussion texts from the Zhihu platform，and adopts a combination of critical incident analysis and grounded theory to conduct a qualitative analysis of public cognition and behavioral changes. Through open coding，axial coding and selective coding，a theoretical model of AI literacy evolution is constructed. Result/conclusion Six core literacy dimensions are extracted：cognition，technology，application，risk，industrial logic，and critical reflection. The evolution of public AI literacy presents a four-stage dynamic process：“cognition formation – practical misalignment – feedback learning – cognition reshaping”. A model of the “dynamic evolution mechanism of public AI literacy driven by technological booms” is further proposed.

Purpose/significance This study aims to investigate the heterogeneity of user behavior in the context of Generative Artificial Intelligence （Gen AI） from the perspective of affective support，revealing users’ diverse motivations，divergent behavior patterns，and their underlying formation mechanisms. Method/process Employing the grounded theory approach，this study conducted a coding analysis of data collected from in-depth interviews and netnography，and constructed the “Context-Driven Evolutionary Model of Gen AI Usage Behavior for Affective Support”. Result/conclusion The findings identify six typical patterns of user behavior： emotional catharsis，self-exploration，social compensation，instrumental facilitation，recreational exploration，and hybrid types. Users interact strategically with general-purpose AI based on their personal experiences and situational contexts，shaping it into a personalized “affective support tool”. This process constitutes the fundamental source of behavioral heterogeneity. The findings elucidate the dynamic procedural mechanisms of human-AI affective interaction and expand the affective dimension of information behavior research.

Purpose/significance Generative artificial intelligence has profoundly transformed the paradigm of information production，while also posing systematic methodological challenges for ensuring content reliability. Method/process This study reviews the research context through bibliometric analysis，based on the perspectives of grand intelligence and information ecosystem governance，taking intelligence thinking as the core methodology. A three-dimensional collaborative framework integrating technological safeguards，process control，and collaborative actors is constructed to achieve full lifecycle management of AIGC trustworthiness. The framework’s explanatory power and adaptability across multiple scenarios are systematically demonstrated through three differentiated cases： the End-Gen project，Doubao AI mobile，and Zhipu Open-AutoGLM. Result/conclusion The three-dimensional framework of technological safeguards，process control，and collaborative actors forms a dynamic cyclical system through mechanisms of thinking empowerment，cross-dimensional linkage，and closed-loop optimization. It not only effectively addresses challenges to AIGC trustworthiness but also enriches the theoretical system of AI content governance in the digital era，providing methodological support for building a trustworthy，safe，and responsible AIGC ecosystem.

Purpose/significance Think tanks can act as bridges to narrow the gap between academic achievements and policy practices. Focusing on the process of knowledge application helps to understand the behavior and function of think tanks in optimizing knowledge processing and influencing policy formulation. Method/process This study takes the policy advocacy of U.S. think tanks on the General Data Protection Regulation （GDPR） as a case and applies the perspective of Actor-Network Theory to systematically examine how think tanks facilitate the transformation of academic knowledge into policies through three stages： knowledge selection，knowledge processing，and knowledge influence. Specifically，the study identifies the characteristics of academic achievements adopted by think tanks，analyzes their patterns of knowledge extraction and the policy intentions underlying the translation process，and traces the adoption and application of their achievements in policy documents. The findings aim to provide insights for improving the translation of academic achievements into policy in China. Result/conclusion The process by which think tanks facilitate the translation of academic knowledge into policy follows a dynamic “selection–translation–mobilization” sequence. Think tanks select academic achievements with evidential value，conduct refinement and contextual reconstruction of them to form policy arguments，and then rely on actor networks to promote the adoption of relevant propositions in policy-making and their phased institutionalization. Based on these findings，this study puts forward implications for scientific researchers，think tanks and policymakers，emphasizing problem-oriented research，strengthening the capacity for policy translation，and improving the regular mechanism for matching knowledge demands，so as to facilitate the effective integration of academic knowledge and public decision-making.

Purpose/significance Current research on papers’ forward citation faces limitations such as the insufficient predictive power of explanatory models and the lack of interpretability in predictive models. This study aims to construct a machine learning model with high predictive performance and interpretable features，reveal the patterns through which lexical characteristics in text semantics influence forward citation，and provide decision aids for academic writing and preliminary journal review. Method/process Using titles，abstracts，authors，and source journals as input variables，the study employed three vectorization methods and six machine learning algorithms to develop a three-category prediction model for the average annual citation counts. We conducted ablation experiments and used SHAP value analysis and partial dependence plots to interpret the specific effects of features on citation frequency，and further designed a word-frequency recommendation method for important terms. Result/conclusion Analysis in the field of new energy vehicle power systems shows that semantic features from titles and abstracts contribute the most to prediction accuracy. Specific hot-topic technical terms help increase citation frequency，while some general technical terms require concise usage. The study further provides specific recommended usage frequencies for particular words. ［Innovation/ limitation This study integrates interpretable machine learning methods with semantic feature analysis，which not only ensures the accuracy of citation prediction but also reveals the specific mechanisms through which keywords in titles and abstracts influence forward citations. However，the model’s findings exhibit domain dependency，and the interaction effects among words have not been thoroughly explored. Future work could analyze the applicability of semantic features across multiple disciplines.

Purpose/significance Faced with intensifying global technological competition and accelerated iteration of disruptive technologies，accurately quantifying competitive dynamics in the multidimensional resource space and identifying differentiated disruptive paths are essential for understanding the mechanisms of technological disruption and guiding forward-looking science and technology strategy. Method/process Using quantum computing patents （2015–2025） as the sample，this study integrates Niche Theory to construct a dynamic heterogeneous network of technology themes and three types of resource nodes. We define four core niche variables—breadth，overlap，density，and shift—to systematically quantify competitive dynamics，and further identify four disruptive paths： functional expansion，scientific pioneering，market development，and cross-domain integration. Result/conclusion Results show that quantum computing exhibits competitive convergence with high niche overlap and density at the scientific and application levels，while the functional level—marked by low density and high breadth—forms the core arena for differentiated competition. The four disruptive paths correspond respectively to dynamic disruption processes driven by functional expansion，knowledge monopolization，market differentiation，or resource integration. This validates the effectiveness of the proposed framework in quantifying competition and identifying disruptive pathways.

Purpose/significance With the rapid development of digital-intelligence technologies and social media，emerging topics，characterized by suddenness，broad reach，and high uncertainty，have significantly influenced the evolution of public opinion and social stability. Accurately identifying emerging topics is therefore of great importance for fostering a healthy online ecosystem. Method/process Taking Weibo trending searches as an example，this study constructs an evaluation framework for emerging topics covering two dimensions，namely Newness and Popularity，with eight core indicators. BERTopic is employed to perform semantic clustering and topic representation on the preprocessed data，and the clustering results are further refined through manual examination. In addition，the CRITIC method is combined with expert judgment to determine indicator weights，thereby enabling the quantitative evaluation and identification of topic clusters. Result/conclusion The results show that topics on online social platforms can be classified into four categories： Emerging （high Newness and high Popularity），New but Not Popular，Popular but Not New，and Neither New nor Popular. Different types of topics exhibit distinct domain distribution patterns，with emerging topics more likely to concentrate in areas involving broad public concern and high sensitivity. Cross-period analysis further indicates that the quadrant distribution of topics remains generally stable over time. Compared with one-dimensional identification based solely on Newness or scale-related Popularity，the proposed approach effectively reduces misclassification and improves interpretability.［Innovation/ value This study proposes a dual-dimensional evaluation perspective based on Newness and Popularity and develops a reproducible identification process. It generates quadrant-based classification results and differentiated governance recommendations that can be linked to monitoring practice，thereby providing methodological support for the early assessment and guided intervention of topics on social platforms.

Purpose/significance Conventional patent quality evaluation approaches rely predominantly on quantitative indicators，such as citation frequency，thereby limiting their capacity to capture the underlying technological semantics embedded in patent texts and the legal significance inherent in citation relationships. Consequently，the accuracy of core patent identification remains restricted. To overcome this limitation，this study proposes a patent quality evaluation method that integrates textual semantic information with the legal attributes of patent citations，with the aim of enabling more precise identification of core patents that exert critical influence on technological evolution. Method/process First，invention patent data from the target technological domain were collected. Following data preprocessing，patent titles and abstracts were extracted and converted into vectorized representations. Subsequently，two key indicators were constructed and quantified：① citation content similarity，calculated using the cosine similarity between the title–abstract vector representations of citing and cited patents；② citation type weight，assigned according to the classification criteria established by the national patent authority，with citation types weighted from 1 to 5 in ascending order of legal and technological significance. Finally，the two indicators were integrated to derive a composite patent quality score，and the top 10% of patents were identified as core patents. Result/conclusion Using invention patents in the Large Language Model domain from 2006 to 2025，a total of 454 core patents were identified，encompassing 72 technological topics and converging into three major clusters： intelligent technology application clusters，data and security technology clusters，and domain-specific technology clusters. Validation results indicate that the proposed method preserves high specificity while demonstrating a substantial enrichment effect for patents with strong commercialization potential. In comparison with conventional citation frequency–based core patent identification approaches，the proposed method achieves superior performance across multiple evaluation metrics，including Precision，Recall，and F1-score. This study not only realizes an effective integration of semantic relevance and technological linkage，but also provides an operational analytical framework for core patent identification within complex technological systems.

Purpose/significance Research on user search behavior in LAM （Library，Archive，and Museum） digital collections has emerged as a core direction for advancing their intelligent transformation and personalized service delivery. Existing studies have primarily focused on the behavioral characteristics associated with traditional topic-based search strategies，while against the backdrop of the widespread application of Artificial Intelligence Generated Content （AIGC） technology，the AIGC-assisted search behavior patterns within digital LAM contexts remain largely unexplored. Method/process This study employed the laboratory-controlled experimental approach. A simulated online LAM search environment was constructed to conduct multiple groups of user experiments. Behavioral differences between traditional topic search and AIGC-assisted search were systematically compared. The behavioral performance of the AIGC-assisted search strategy across multiple task types was further examined，and key search behavior indicators for distinguishing task categories were identified. Result/conclusion Significant differences were observed between the two search strategies in behavioral characteristics such as information browsing and question-answer interaction； When the AIGC-assisted search strategy was adopted，significant behavioral variations also emerged across different task types，and query-related behaviors exhibited the strongest discriminative power for task classification. These findings not only reveal the considerable potential of AIGC technology for enhancing human-machine collaborative search efficiency in the context of LAM digital collection retrieval，deepening the understanding of the inherent mechanism of human-machine integrated searching behavior，but also provide theoretical support and practical guidance for optimizing intelligent search service systems in LAM environments oriented toward complex cognitive tasks.

Purpose/significance To enhance the discoverability of scientific datasets and strengthen users’ trust in them，effective methods are needed to identify dataset-related literature that provide users with data descriptions and usage contexts. Method/process This study obtains related literature based on citation texts found in the metadata of dataset. First，a large language model for entity recognition is trained to extract entities such as the title，author，year，and source from citation texts. Second，an edit-distance algorithm is used to mitigate hallucinations in entity recognition. Third，the recognized titles are used as queries to search literature databases and obtain candidate publications. Finally，a large language model for literature scoring is trained to select the best match from the candidate publications. Result/conclusion Metadata of 478，481 datasets were collected from 111 scientific data repositories to construct an evaluation dataset and conduct experiments. The results show that the large language model–based recognition method achieves the best performance： the citation entity recognition model reaches an F1 score of 0.9737，and the literature matching scoring model achieves a weighted matching accuracy of 0.9705. ［Innovation/ value Large language models are used to improve the quality of citation entity recognition and literature matching. An edit-distance algorithm is introduced to alleviate hallucinations in entity recognition. By combining the proposed method with other approaches （e.g. DOI matching），dataset-related literature can be retrieved more comprehensively.

Purpose/significance To address the issue that false information in aerospace and aviation science and technology news is often disseminated in multimodal forms，where single-text analysis is insufficient for effective verification，this study explores the use of key aircraft objects in news images as visual evidence to support intelligence analysis. Method/process A dedicated aerospace and aviation science and technology news object detection dataset is constructed. Based on the Deformable DETR framework，we propose an MS-SOAR module that enhances object representation capability through cross-scale feature enhancement and a semantic-guided multi-level feature fusion mechanism. Result/conclusion The proposed method significantly outperforms baseline models in overall object detection accuracy and achieves more balanced performance across objects of different scales. These results demonstrate the effectiveness of the proposed cross-scale feature enhancement and semantic-guided representation learning strategy，providing reliable visual support for misinformation identification in science and technology news.

Purpose/significance To address insufficient interaction efficiency and reasoning capability in intelligent diagnostic guidance systems，this paper proposes an interactive guidance system based on a weighted knowledge graph，providing a methodological pathway for informatics-supported personal health decision-making. Method/process A weighted knowledge graph is constructed by integrating medical knowledge graphs with real consultation data. Candidate diseases are located via dynamic subgraph sampling，and optimal symptoms are selected for inquiry using Bayesian inference and a maximum discrimination strategy，achieving accurate guidance with minimal interaction rounds. Result/conclusion On the MZ，DXY，and GMD datasets，disease guidance accuracies reach 0.75，0.84，and 0.76，improving over state-of-the-art methods by 1%，2%，and 3%； symptom inquiry efficiencies reach 0.27，0.30，and 0.40，improving by 11%，6%，and 15%，validating the method’s effectiveness in accuracy，efficiency，and interpretability.

Purpose/significance Generative AI （GAI） hallucinations have emerged as an inevitable byproduct of the rapid development of large language models. Existing studies have predominantly addressed hallucinations from a technical perspective，emphasizing model optimization and detection tools. However，the emergence and governance of hallucinations cannot be reduced to a purely technical problem； rather，they are fundamentally socio-technical issues deeply embedded in sociocultural contexts and shaped by users’ cognitive processes. This paper aims to deepen the theoretical understanding of GAI hallucinations as a complex social phenomenon by synthesizing and analyzing the main research trajectories and core themes in the existing literature，thereby offering insights for mitigating hallucinations and developing trustworthy AI. Method/process Drawing on socio-technical systems theory，this study adopts a hermeneutic literature review to examine the conceptual foundations of GAI hallucinations systematically，the pathways through which hallucinations influence human-AI interaction，and frameworks for multi-stakeholder collaborative governance. Result/conclusion In the context of GAI hallucinations，the user behavior chain encompasses need articulation，evaluation，cognitive reconstruction，and behavioral consequences. Meanwhile，hallucination governance exhibits the characteristics of a complex socio-technical system involving multi-stakeholder participation and cross-level coordination. Future research may further explore core issues such as multimodal feature identification of GAI hallucinations，human-AI interaction patterns，strategies for enhancing user literacy，and agile governance mechanisms. ［Innovation/ value By deconstructing GAI hallucinations through the lens of socio-technical systems theory，this study deepens a systematic understanding of this phenomenon，clarifies the links between this social phenomenon and relevant fields，and also provides theoretical insights for building a trustworthy GAI ecosystem characterized by multi-stakeholder collaborations and agile responsiveness.