The rapid rise of generative artificial intelligence is propelling digital humanities (DH) toward a paradigm shift to AI-driven intelligent humanities (AI-DH). This study provides a systematic account of this transformation by examining its external morphological evolution, internal methodological innovations, and underlying cognitive reconstruction. In terms of research morphology, AI-DH is characterized by human-AI collaboration, integrated and generalized technological architectures, the activation of data intelligence, and the emergence of novel research forms. Cognitively,AI-DH embodies the dialectical unity of human and machine intelligence, the convergence of humanistic rationality and probabilistic inference, and the coexistence of deep exploration and broad generalization. Methodologically, AI-DH demonstrates a theoretical shift from tool-based assistance to intelligent partnership, from humanities-centered to humancentered inquiry orientation and a transition from linear workflows to open generative systems. These shifts give rise to four emerging paradigm patterns: generative humanities, dialogic humanities, simulated humanities, and extended humanities. Overall, AI-DH not only expands the boundaries and capabilities of traditional DH but also reshapes its epistemic foundations and practical pathways, offering a systematic framework for the intelligent evolution of humanities research in the era of generative AI.

This paper systematically analyzes the core challenges faced by artificial intelligence technologies, particularly large language models (LLMs), in museum applications. By examining the structural limitations inherent in LLMs, it further explores the practical difficulties encountered by museums when implementing AI. In response to these issues,systematic countermeasures and a tiered governance framework are proposed to provide methodological support for the controlled implementation of AI in the cultural heritage sector. The study demonstrates that through prudent planning and scientific governance, artificial intelligence can serve as an effective tool to enhance the core operational capabilities of museums, offering sustainable momentum for innovation in cultural heritage preservation, research, and public services.

从2016年到2025年，上海图书馆开放数据竞赛已连续举办十年。作为国内公共文化领域最早探索向社会开放人文数据的实践之一，竞赛从单一机构以家谱专题资源为起点，逐步拓展至多家机构共同参与、涵盖多类数据主题的开放数据生态。十年来，竞赛持续推动开放数据体系建设，形成了“政府指导、图书馆主导、社会协同、企业助力”的合作格局，以及集“竞赛、宣讲、培训、展览和分享会”“五位一体”的办赛模式。竞赛以文化与科技融合的力量，让图书馆的数据“活”起来，让数据真正成为社会的新质生力，推动图书馆的角色从知识的“仓储”转变为智慧的“孵化器”。

By conducting a systematic study of China’s artificial intelligence governance policies through thematic modeling and quantitative evaluation, this paper provides support and guidance for policy optimization. Effective data from the 92 nationallevel artificial intelligence governance policies issued in China from 2017 to the 15th of March, 2025 were preprocessed.The BERTopic model was used to identify policy themes, calculate keyword c-TF-IDF weights, and three other thematic indicators, and combined with policy content to conduct an in-depth analysis of governance policy priorities. Based on thematic modeling, the PMC policy evaluation model was constructed to conduct empirical quantitative analysis on nine representative policy documents. The BERTopic model identified 13 themes, covering nine key areas of China’s AI governance policies. The PMC index model indicates that the overall policy level is good, but further optimization is needed in terms of governance framework, governance tools, and governance entities. Based on this, policy recommendations were proposed, including the construction of a classified governance policy framework, the combination of policy tools, and the promotion of collaborative governance, to serve China’s AI governance practices.

In studies on migration and remittances, the impact of kinship strength on international remittances lacks empirical testing. This paper constructs a remittance database of Chaozhou-Shantou immigrants in Southeast Asia from 1915 to 2000 based on the Chaozhou-Shantou Overseas Correspondence archives. Within the framework of digital humanities, it reconstructs kinship structures based on the titles used in the correspondence, quantifies transnational kinship indices, and, combined with clan associations, hometown associations, Chinese-language newspapers, and genealogical records, constructs social network indicators. Ordinary least squares and two-way fixed effects estimates show that the recipient’s kinship intensity index is the most robust predictor of remittance behavior: each increase in kinship rank is associated with an increase of about 20% in remittance frequency relative to the sample mean. Once sending and receiving locations are held constant, the explanatory power of associational and media networks declines, and Qiaopi agencies function mainly as transaction channels rather than sources of motivation. Interaction terms indicate that genealogical documentation amplifies the effect of kinship,suggesting that lineage institutions strengthen family-based remittance obligations through collective memory and reputational constraint. The study integrates historical textual sources with econometric methods and provides Chinabased evidence on remittances from a social network perspective.

With the rapid advancement of information technology, the development of smart libraries has become a core task and strategic direction for the transformation and development of the global library field. This paper,grounded in the context of smart library development being elevated to a national strategy during the 14th Five-Year Plan period, delves into the shift in smart library evaluation paradigms—from performance evaluation focused on efficiency to effectiveness evaluation emphasizing tangible outcomes. It systematically reviews the development of public library evaluation in China and its role in driving technological adoption. It distinguishes between performance evaluation and effectiveness evaluation by examining their core concepts, theoretical foundations, and international practices. Furthermore, it constructs a comprehensive evaluation framework encompassing multiple dimensions:smart spaces, smart services, smart resources, smart management, and smart librarians. Drawing on representative domestic and international case studies, this paper analyzes the key technologies, practical achievements, and challenges currently facing smart library development. It aims to provide theoretical references and practical pathways for the scientific evaluation and sustainable development of future smart libraries, ultimately driving the professional advancement of library services through management empowerment. This will enable libraries to demonstrate their irreplaceable social value in the era of digital intelligence.

In the context of national strategies for educational empowerment and digital China initiatives, the development of smart libraries has become a pivotal component of the national cultural digitalization strategy, necessitating artificial intelligence (AI) to drive service model innovation. This paper focuses on university library circulation and reference service scenarios, guided by the practical needs of faculty and student users. A general-purpose agent-based intelligent question-answering system was developed using a knowledge base of frequently asked questions (FAQs) in circulation operations. By integrating AI large language model to achieve automated knowledge reasoning and response generation, the system significantly enhances service efficiency while optimizing limited human resources. Empirical results demonstrate that this solution markedly improves library service effectiveness, provides a reusable technical framework for intelligent transformation in other domains, and contributes to advancing digital transformation and modernization in higher education.

The profound transformation of the knowledge ecosystem and technological environment, along with the deepening evolution of library services, has driven continuous advancements in the mission of librarians. The new missions of libraries include roles such as literacy cultivators, research collaborators, community connectors, experience designers,digital literature preservers, and advocates of open science. This article will focus on three key directions—digital transformation, multidimensional literacy, and open science—to connect the future development of libraries from the perspective of these new missions. Aligning with the phased goals for 2035, it proposes that libraries should become learning hubs in the digital society, spaces for cultivating civic literacy, and intermediary platforms for open science. Through phased planning of action routes, it envisions how librarians can shape a new paradigm of knowledge sharing with professional dedication and social responsibility in a future where intelligence and humanity coexist.

Artificial intelligence (AI) is reshaping scientific research paradigms at an unprecedented speed and breadth.From protein structure prediction to assisting drug development, AI technology has become an important tool in scientific research activities. However, current AI still has significant limitations in reasoning ability, knowledge generation, and experimental design, and cannot fully replace the role of human scientists. As a part of the knowledge infrastructure, the library must reposition its function and value in the scientific research ecosystem. Based on the current development trend of AI scientific applications, this article analyzes its impact and challenges on libraries, and proposes five key action strategies for the future to achieve the transformation from an “information service provider” to a “research collaborator”.

This study tackles the issues of lack of narrative appeal, low engagement and poor historical authenticity in digitizing red culture resources, by proposing an interdisciplinary framework for red-themed educational games, integrating self-determination theory and flow theory to address adolescent psychological needs. The approach employs retrievalaugmented generation (RAG) and a dual-layer prompt mechanism to dynamically retrieve knowledge and ensure content accuracy, ensuring historical authenticity, and effectively enhancing the interest and immersive experience of red education.A game prototype, Retracing the Red Journey, was developed using data from the Shanghai Library and Jiaxing Library.Results confirm its effectiveness in enhancing user immersion, engagement, and historical understanding, offering a practical model for innovating red culture education.

This study targets the national security intelligence analysis scenario, aiming to overcome the informational limitations of single-modal data analysis and enhance situational awareness in complex scenarios. Based on deep learning theory, this paper constructs a heterogeneous data representation model from the macro perspective of multimodal data fusion, organizes a cross-modal data fusion framework, and introduces a dynamic weight allocation mechanism to achieve effective integration of multi-source heterogeneous data such as social media text, video, audio, satellite imagery, and network traffic. The research indicates that, in a complex data ecosystem, attention should be paid not only to the granularity of multimodal data fusion in open-source intelligence, but also to the requirements of intelligence tasks, enabling rapid and accurate fusion to form a practical and feasible industrialized process.

The proliferation of Artificial Intelligence Generated Content (AIGC) is fundamentally challenging the ontological foundations and managerial logic of Information Resource Management (IRM). Traditional IRM paradigms, predicated on a static lifecycle, prove inadequate for the extreme dynamism, emergent collaboration, and heterogeneous complexity of the AI open-source ecosystem. This paper, through an analysis of communities like Hugging Face and domestic models such as DeepSeek and Qwen, identifies three critical dimensions of generative information resources: Throughput (T),Quality (Q), and Capability Ceiling (C). Integrating insights from established theories, this paper constructs the dynamic Information Resource Management Theory (IRMT), a novel framework featuring the T-Q-C trade-off triangle to guide the strategic optimization of AI resources. This work offers more than a new theoretical paradigm for an autonomous IRM knowledge system in the AIGC era; it also delineates a practical path toward achieving efficient application of domestic AI and technological self-reliance, especially under computational constraints. Future research avenues include the quantitative modeling of the T-Q-C framework, the integration of ethical governance, and the extension of the theory to multimodal and agent-based systems.

This study focuses on the activities and practices of Lu Xun during his Shanghai residence period (1927—1936)as the primary research object, with the aim of systematically revealing his behavioral strategies and intellectual trajectory within a specific historical context through digital methods. First, by integrating linked data and narrative theory, a digital storytelling model was constructed to support semantic associations and spatiotemporal visualization. Second, resources such as the Shanghai Library API and Lu Xun’s chronology were systematically integrated and organized, on which basis a digital storytelling platform was developed to realize the narrative presentation of Lu Xun's life trajectory and Shanghai’s urban space in spatiotemporal dimensions, followed by an analysis thereof. This study achieves visualized narratives based on spatiotemporal semantics, provides a verifiable digital storytelling paradigm for research on Lu Xun’s later thoughts, and expands the methods of digital humanities in the narrative analysis of historical literature.

The ethical challenges posed by artificial intelligence (AI) have spurred the development of numerous assessment scales for AI ethical literacy. However, widespread methodological flaws in these scales are fueling a “reproducibility crisis”, which has become a key bottleneck hindering the implementation of ethical education and governance practices. This study adopts the internationally recognized COSMIN guidelines. By searching Chinese and international databases for literature published between 2015 and 2025, 20 AI ethical literacy scales were selected and included for a systematic evaluation of their methodological quality. The results show that the existing scales are geographically widespread, with China having the most (7 scales). The assessed populations cover diverse groups, including algorithm engineers, students,and the general public, and the primary rating format is the Likert 5-point scale. The assessments focus on dimensions of ethical cognition, attitude, and behavior. While the scales perform well on metrics such as internal consistency and structural validity, issues such as non-standardized development processes, insufficient evidence supporting content validity, and a lack of cross-cultural applicability are prevalent. Based on the evaluation criteria of the COSMIN guidelines, four scales ultimately received an A-grade recommendation, with their application scenarios focusing on ethical literacy assessment within specific fields such as healthcare and education. This study provides an empirical foundation for the scientific selection and optimized enhancement of assessment tools for AI ethical literacy.

To understand the formation of academic figures’ relationships in the Song Dynasty and explore the integration path between digital humanities technology and traditional historical research, a Song Dynasty academic network involving 1 987 scholars and 2 742 groups of relationships was constructed based on the china biographical database(CBDB). The research focused on three core issues: the impact of scholars’ attributes (hometown, school identity,etc.) , social capital (centrality, structural holes, etc.) , and the endogenous structure of the network (reciprocity,transitivity, etc.) on the establishment of academic relationships. Geographic proximity and homogeneity of school identity have a significant positive effect on the formation of academic connections. The network shows strong reciprocity but suppressed transitivity, reflecting the bidirectionality of academic interactions and the competitive exclusion between schools. In terms of social capital, centrality, especially degree centrality and eigenvector centrality,has a significant impact on relationship building, while the role of structural holes is limited. This study provides empirical evidence for understanding the social foundation of academic dissemination in the Song Dynasty and offers a new path for the integration of digital humanities and traditional historiography.

This paper focuses on the construction of a knowledge graph for the chronological resources of Xian Xinghai.Based on Qin Qiming’ Brief Chronicle of Xian Xinghai (1905—1945), an ontological knowledge model was developed,encompassing core entities such as persons, time, locations, events, works, and documentary sources. The Neo4j graph database was employed for data storage and relational querying, while the Vis.js library facilitated the dynamic visualization of Xian Xinghai’s social network and life trajectory. The results demonstrated that knowledge graph technology can effectively enhance the analyzability and visualization of Xian Xinghai’s life events, social networks, and artistic creation context, overcoming the limitations of traditional linear chronological representations and providing new research methods and pathways for understanding his music and revolutionary practices across different regions and stages. Future work may focus on enhancing the efficiency of knowledge extraction, expand data sources and enhance the interactive visualization experience.

In the “screen-reading era”, user demand for knowledge content in media formats such as video, audio, and images has surged dramatically. During the 15th Five-Year Plan period, libraries are undergoing a critical transition from intelligent to smart operations. Within the macro framework of building high-quality data resources, establishing a comprehensive media knowledge organization system, efficiently managing media resources, and unlocking their new value to provide users with diverse knowledge access and service support has become particularly crucial. This paper examines the case study of Guangzhou Library’s smart media resource management and service platform development. Addressing current challenges in library media resource management and application, it summarizes experiences and outcomes in resolving these pain points, and proposes further concepts for the 15th Five-Year Plan period. Further envisions how to deepen and refine media asset intelligent applications from three dimensions—“enhancing resource utilization, enhance AI capabilities, and building a humanistic hub”—to unlock the platform’s potential value in public cultural services and academic research support. This aims to provide new pathways for consideration in the construction and application of high-quality data resources for libraries.

The deepening development of digitization and intelligence has generated a huge demand for digital intelligent talents. In the array of disciplines and majors in universities that cultivate digital intelligent talents, the major of information management and information system occupies an important position. This study contributes to the theoretical development of the academic discourse on cultivating digital intelligent talents in the field of information management and information system and responds to the needs and challenges posed by the era of digital intelligence at a practical level. Based on the definition of the competency of digital intelligence talents for the major of information management and information system, we collect data on digital intelligence-related job postings for undergraduates in information management and information system from a large online job search website. It then extracts relevant categories using grounded coding method, ultimately developing a demand-driven competency model of digital intelligence talents for the major of information management and information system. The model identifies 13 main categories, classified into three groups: knowledge, skills, and personal qualities, with intrinsic logical relationships among them. The model shows that digital intelligence has reconfigured the traditional core competencies of talents in information management and information system, further expanding the competency structure of talents in the context of big data, giving more prominence to development skills,and encompassing a broader range of new skills.

In the context of the marketisation of data elements, revealing the integration mechanism between scientific data and the various elements of new quality productive forces can provide theoretical references and practical guidance for promoting the value transformation of scientific data and accelerating the development of new quality productive forces. Following the “path dimension-implementation path” framework, this paper analyses the path dimension of scientific data empowering the formation of new quality productive Forces from three horizontal dimensions: technology,institutions, and ecology. It then proposed specific implementation paths for scientific data to empower new quality productive forces from three vertical dimensions: new-type workers, new-type means of production, and new-type objects of labour. Scientific data empowers the formation of new quality productive forces through three pathways: driving innovation in research paradigms, facilitating the market-based allocation of data elements, and constructing cross-subject collaborative innovation networks. This can be achieved by enhancing the data literacy of new-quality workers, implementing legal governance of scientific data, and endowing traditional labour objects with new characteristics to empower new-quality workers, new-quality labour materials, and new-quality labour objects through scientific data.

Data, as a new factor of production, raises new requirements for talent cultivation in the field of information resource management (IRM). The undergraduate curriculum system, as an important stage of talent cultivation, is a key breakthrough for the IRM discipline to move towards the field of data factor. Through literature review and text analysis,the current status of the undergraduate curriculum system in IRM was systematically reviewed. The study found that the current undergraduate curriculum system is gradually formed by stacking in sequence , making it difficult to flexibly adapt to national policy requirements and the direction of disciplinary reform. Data has an inherent affinity with the IRM discipline and serves as a bridge connecting various domains. In the background of data factorization, data acts as a key lever for integrating the IRM curriculum system, facilitating the formation of a tiered talent development model. Based on this,integrating the knowledge system of the IRM discipline, this paper proposes an integrated restructuring strategy based on data: firstly, at the level of general platform courses, existing data knowledge units are shifted downward and integrated, with data as the foundation, technology and management being equally emphasized, and the basic theories of disciplines being optimized synchronously; secondly, at the level of core major courses, data elements are deeply embedded to form an “theory-technology-management” integrated curriculum framework. This new curriculum system not only facilitates dynamic adjustments in undergraduate majors of information resource management but also meets the academic research needs of the discipline, cultivating excellent talents for the flexible job market and further academic advancement, so as to better serve economic and social development.

The rapid development of generative artificial intelligence (GAI) presents a strategic opportunity for the intelligent transformation of libraries. However, there is a clear disconnect between current research and practice at the project decision-making level: libraries generally lack meso-level analytical tools to assess the feasibility and strategic fit of AI projects, leading to fragmented innovation paths, resource misallocation, and strategic anxiety. Focusing on academic library settings, this study constructs an AI innovation project evaluation framework that encompasses five dimensions:technical difficulty, cost threshold, compliance risk, user value, and strategic window. On the basis of clarifying the definitions and assessment content of each dimension, the framework is applied to two typical cases—“AI-based scholarly search” and “AI literacy education”—to systematically demonstrate its practical use in project analysis and decision trade-offs, and to summarize a phased, iterative evaluation process and project categorization strategy. The findings show that this framework can help libraries identify quick-win projects, strategic projects, and cautious projects among multiple options, thus enabling a shift in decision-making from “whether to introduce AI” to “how to select and implement AI in a scientifically grounded way”. The results provide theoretical support and practical guidance for libraries to build innovation pathways that are strategically coherent, risk-controllable, and value-sustainable in the course of intelligent transformation.

To explore the diagnostic and treatment rules and knowledge in ancient books, and enhance their utilization rate and academic value, this study, relying on the collection of ancient books in the Library of Shanghai University of Traditional Chinese Medicine, constructed multiple AI agents for ancient medical records by utilizing natural language processing technology, deep learning technology, and multi-agent cluster technology. These agents include: the content extraction agent for ancient book sorting, which is responsible for automatic indexing, structuring, and vectorization of ancient book texts; the health consultation assistant agent based on ancient medical records, which provides personalized health advice for users based on ancient medical records; and the traditional Chinese medicine (TCM) popular science knowledge Q&A agent, which offers Q&A services on TCM knowledge and basic library business consultation. Through the collaborative operation of the above agents, the project has achieved good results, realizing a full-chain intelligent service from the sorting of ancient medical records to clinical assistance and cultural dissemination, and providing an innovative solution for the inheritance and development of TCM.

This article reviews four types of reading promotion methods and their limitations, analyzes the characteristics of AI podcasts including dual-host dialogue, natural voice synthesis, and human-computer interaction potential, and reveals the mechanisms by which they lower reading barriers through colloquial explanations, fragmented reading, and personalized options. The study indicates that AI podcasts have important significance in enriching the reading ecosystem,innovating promotion strategies, and facilitating human-machine collaboration. Meanwhile, it proposes response strategies such as quality review and legal authorization to address issues of content accuracy, copyright protection, and ethical transparency, emphasizing the need to rationally view AI podcasts as tools while leveraging the irreplaceable role of human promoters.

In the context of a research environment deeply empowered by generative artificial intelligence, from the perspective of building an intelligent matching mechanism, this paper aims to promote the collaborative innovation between industry, academia, and research, as well as to enhance industrial application capabilities. This paper primarily relies on a large language model and retrieval-augmented generation (RAG) framework, with multi-dimensional user and expert profile modeling and semantic alignment methods as the foundation. Additionally, it introduces a heterogeneous data processing mechanism, multi-dimensional ranking algorithms, and an explainable recommendation framework. The proposed end-to-end expert recommendation system has achieved significant breakthroughs in improving matching accuracy, result explainability, and system adaptability, providing both a theoretical foundation and practical support for enhancing collaboration efficiency between enterprises and academia.

Traditional manual cataloging models are facing sustainability bottlenecks amid the explosive growth in the scale and diversity of information resources, becoming a critical constraint on library knowledge services. While generative artificial intelligence offers new technological opportunities for automated metadata creation, existing explorations often suffer from insufficient workflow integration, difficulties in quality control, and weak adaptability across different document types. To systematically address these challenges, this paper proposes an integrated intelligent cataloging framework for multi-type documents, incorporating fragmented AI-based cataloging practices into a scalable and implementable system. The framework provides a structured methodological foundation for the systematic adoption of large language models in library environments.Beyond proposing technical solutions adaptable to diverse document types, the framework also tackles the challenge of aligning traditional cataloging rules with AI-driven processes by introducing a dual-path mechanism of MAC and OAC, thereby resolving structural conflicts between AI technologies and cataloging standards. Finally, through case studies involving special collections on Chinese traditional opera playbills and books, the paper demonstrates the effectiveness and scalability of the proposed framework, offering both a theoretical basis and a practical paradigm for AI-driven, high-quality metadata production.

With the continued advancement of China’s “Smart Museum” strategy, smart guide systems, virtual digital human, and generative AI are becoming key infrastructures in museums. Drawing on the learning ecology framework, this study conceptualizes museums as key nodes in young people’s personal learning networks and defines smart applications as technical devices for reconfiguring such nodes. It proposes four analytical dimensions—situational interfaces, resource systems, relational mechanisms, and opportunity pathways—and compares representative domestic practices. Taking the “Smart Guide” WeChat mini program of the Shanghai Science and Technology Museum as a case, the study examines how a unified digital front desk, exhibit-centered resource organization, the extension of support through the AI assistant and family practice, and the linkage of recommendations and activity entry points together reshape the ways in which museum nodes are accessed, organized, and connected across contexts.

This paper addresses the widespread challenge of “hidden collections” in library special collections in the digital age, where a vast number of precious materials remain inaccessible to academia due to inadequate cataloging. Traditional manual cataloging is inefficient, while optical character recognition (OCR) technology has inherent limitations when dealing with historical documents featuring complex layouts and diverse fonts. To overcome this bottleneck, this study develops and validates a novel intelligent cataloging paradigm based on a multimodal large language model (MLLM). Its core is a “VLM + LLM + HITL” workflow that integrates a vision-language models (VLM), a large language models (LLM), and a humanin-the- loop (HITL) mechanism. Using special collections on Chinese traditional opera playbills with extremely complex layouts as a case study, this research conducts a comprehensive quantitative evaluation across four dimensions: efficiency, accuracy,cost-effectiveness, and user experience. Experimental results show that this paradigm reduces the average processing time per document from one hour to 20 minutes, achieving a 66.7% efficiency gain. Building upon an initial AI accuracy rate of75%, near-perfect final accuracy (approaching 100%) is attainable through human-AI collaboration. Cost-effectiveness analysis reveals a benefit-cost ratio as high as 2.78. Furthermore, a user experience evaluation involving professional catalogers yielded a high system usability scale (SUS) score of 82.5. The core contributions of this study are threefold: Technically, it demonstrates the superiority of VLMs in intelligently processing complex documents. Methodologically, it proposes a multidimensional evaluation framework encompassing error type analysis, cost-effectiveness, and user experience. Practically, it provides a validated, scalable solution with economic feasibility and favorable user acceptance for addressing the “hidden collections” problem and revitalizing massive special collections, thereby offering crucial technical and methodological support for the deeper implementation of national cultural heritage digitization strategies.

AI literacy exhibits distinct characteristics of interdisciplinary integration, with the rich resource endowments of libraries serving as the core foundation for its educational implementation. Through a systematic analysis of theoretical frameworks and typical cases of AI literacy instruction, this study finds that AI literacy is not a static list of competencies and a collection of knowledge, but rather an open system that is continuously reconstructed alongside technological evolution. Leveraging the resource advantages of libraries, strengthening top-level design, and constructing a systematic educational framework within this dynamic evolution are pivotal to the standardized development of AI literacy education in libraries.

Promoting the separation of ownership and use rights reform in public cultural facilities is a key driver for energizing these institutions and propelling high-quality public cultural services. In theory, the two-rights separation reform keeps ownership and the public-interest nature of facilities intact while transferring—wholly or partially—the right to operate them, thereby raising quality and efficiency. In doing so, it helps to solve the operational dilemmas of grassroots publiccultural facilities, incubate high-caliber operators, expand inclusive yet non-basic public-cultural services, upgrade government capacity for cultural governance, and generate new employment. To keep the two-rights separation reform on track, a four-in-one linkage mechanism should be built that sets standards for entry and exit, norms for operation and development,safeguards for participation and investment, and feedback loops for performance. In practice, the reform should be rolled out first in five priority fields: 1.extending opening hours to raise service efficiency; 2.sending high-quality cultural resources directly to grassroots communities to boost public satisfaction; 3.accelerating high-value cultural-creative product development to amplify cultural outreach; 4.building public-cultural community that mobilize whole-of-society participation;and 5. spurring internal reform of public institutions to sharpen their vitality.

In-depth understanding of grassroots public cultural needs is the key to achieve high-quality cultural resources directly to the grassroots. Urban communities are the basic units of grassroots public cultural services. According to community public cultural needs, important levels of resource allocation are divided, and the supply mode and content of cultural resources are optimized to provide accurate allocation schemes for solving the “last mile” problem of public cultural services. Interviews and literature analysis were used to extract and summarize the actual cultural needs of the public in urban communities. The KANO questionnaire was designed based on 34 cultural needs and the survey data were analyzed. The types of cultural needs of the public in communities were defined by the Better-Worse coefficient. It puts forward the accurate allocation strategy of high-quality cultural resources in urban communities, which is to continuously cultivate the necessary demand allocation, focus on the expectation demand satisfaction, accurately locate the difference of attractive demand, and continuously track the evolution of undifferentiated demand.

With the rise of generative artificial intelligence, represented by large language models (LLM), library reference services are undergoing a profound paradigm shift. However, traditional evaluation methods face an “evaluation gap” in this context, struggling to measure the factuality, relevance, and overall value of generated content. To address this issue, this study aims to construct a comprehensive evaluation system to scientifically and holistically assess the effectiveness of LLM-based reference services. Grounded in Professor Ye Jiyuan’s “All-round academic evaluation system” framework, the research creatively integrates cutting-edge evaluation metrics from retrieval-augmented generation (RAG) systems to build a threedimensional indicator framework encompassing “formal evaluation” (system technical quality), “content evaluation”(core information quality), and “utility evaluation” (real-world value). The system advocates for establishing a pluralistic and collaborative evaluation ecosystem involving users, librarians, developers, and “AI evaluators” (LLM-as-a-Judge), and promotes the adoption of a mixed-methods approach combining quantitative and qualitative methodologies. Through a practical exploration using the Shanghai Library’s AI Reader Q&A Assistant (hereinafter referred to as “AI Assistant”) as a case study, this research elaborates on the application process of the evaluation system in both formative (guiding optimization) and summative (supporting decision-making) evaluations, thereby verifying its feasibility. The findings of this research aim to move beyond assessing mere “system correctness” towards a comprehensive measurement of the “holistic value” of the service. It provides the library community with a systematic solution that has both theoretical depth and practical significance for evaluating, governing, and developing such emerging intelligent services.

With the in-depth advancement of the third-generation artificial intelligence toward cognitive intelligence,the integration of knowledge graphs and retrieval-augmented generation (RAG) technology has emerged as a core approach to upgrading the intelligent level of library information services. Aiming at the key issues including low efficiency and static limitations of traditional knowledge graphs, as well as the “hallucination” problem of large language model (LLM) and the insufficient multi-hop reasoning capability of conventional RAG, this paper proposes a dynamically collaborative library knowledge model based on LLM and GraphRAG. Innovatively, this model integrates the advanced natural language deep comprehension capability of LLM with the advantages of structured representation of knowledge graphs, enabling the dynamic construction and incremental update of academic knowledge graphs in a crowd sensing environment. By fusing the dual retrieval capabilities of graph databases and vector databases, it constructs a multigranular and interpretable GraphRAG retrieval-augmented mechanism, which significantly improves the accuracy and logicality of academic information retrieval and knowledge-based question answering. This model has been practically applied to the AI-based smart knowledge service platform. Corresponding knowledge service optimization modes are designed to enable intelligent retrieval and intelligent Q&A of academic resources. Empirical research shows that the system possesses good usability, effectively addressing pain points in library knowledge services such as frag mentedretrieval and unreliable Q&A responses. On this basis, an optimized knowledge service model is further designed.The research provides a theoretical model and practical solution for the construction and application of library knowledge graphs in a crowd sensing environment, and further enhances the application efficiency of the new-generation artificial intelligence in the scenario of professional library information services.

As LLM evolve from generative AI toward AI agents capable of independent planning and action, the governance challenges posed by their autonomy—such as blurred accountability and uncertain risks—have become increasingly prominent. Taking EU and US regulations on AI agents as its research object and adopting a comparative analysis method,this paper first defines the core characteristics of AI agents from three dimensions: action autonomy, risk uncertainty, and liability ambiguity, and provides a three dimensional analytical framework covering “boundaries of intervention capacity,timing of regulatory intervention, and rules of responsibility allocation”. On this basis, it systematically compares the EU’s risk‑based, ex‑ante preventive model, with the US’s innovation‑friendly, ex‑post approach. The findings show that while existing EU and US regulations can provide some effective oversight of AI agents, they face growing inadequacies: the EU model suffers from a “human‑supervision paradox”, while the US model encounters a “black‑box accountability problem” in multi‑agent collaborative scenarios. Therefore, this paper proposes that China should absorb the legal certainty of the EU model and the regulatory flexibility of the US model, pursuing a middle path characterized by tiered classification,clear accountability, traceability, and dynamic adaptability, so as to provide theoretical guidance for AI governance in the age of agents.

Against the backdrop of the global digital wave, the transformation of national libraries has evolved from technological upgrades into a systematic project concerning strategy, culture and national competitiveness. Taking the Library of Congress, the British Library and the Bibliothèque nationale de France as research objects, this paper deeply analyzes the distinct models formed by the three in digital transformation: the technology-led and strategy-driven model of the United States, the partnership network and diversified financing model of the United Kingdom, and the state-led and digital sovereignty model of France. The paper makes a systematic comparison from multiple dimensions such as governance structure, strategic system, technological innovation, legal protection and social effects, and extracts their core paths and key measures. Based on the strategic context of the “Digital China” initiative, this paper further explores the reference value of the experience of the three countries for the Chinese libraries. It proposes to build a governance system featuring “state leadership and top-level drive”, explore an empowerment path of “standard guidance and ecological coordination”, implement the infrastructure principle of “technological independence and sovereign control”, and deepen the public cultural supply of “smart services and value creation”, so as to explore a transformation path with Chinese characteristics that integrates strategic determination, technological leadership and ecological collaboration.

“Neck Stuck”key core technologies are the critical constraints on a country’s industrial security and technological competitiveness. Constructing an operable and accurate identification method for “Neck Stuck” key core technologies can provide important decision support for breaking through technological blockades and achieving high-level independent development of science and technology, and has strategic value for safeguarding national economic sovereignty and enhancing international discourse power. Taking the field of lithography technology as an example, this study selects 6 indicators from the perspectives of complex innovation, technological monopolization and core competitiveness to construct an indicator system for identifying key core technologies, based on the connotative characteristics of key core technologies.At the same time, Altmetrics data is introduced. On the basis of identifying the key core technologies in this field, the “Neck Stuck” key core technologies are further screened out from three dimensions: technological gap power, breakthrough difficulty power and social attention power. Using the patent data of the lithography technology field from 2005 to 2024 as samples, the study finally clarifies the distribution of “Neck Stuck” technologies and advantageous technologies in this field in China. This result is highly consistent with the current development trend of lithography technology, which also verifies the scientificity and effectiveness of the proposed three-force identification model for “Neck Stuck” key core technologies in terms of theoretical construction and practical application.

With the rapid advancement of information technology, the development of smart libraries has become a core task and strategic direction for the transformation and development of the global library field. This paper,grounded in the context of smart library development being elevated to a national strategy during the 14th Five-Year Plan period, delves into the shift in smart library evaluation paradigms—from performance evaluation focused on efficiency to effectiveness evaluation emphasizing tangible outcomes. It systematically reviews the development of public library evaluation in China and its role in driving technological adoption. It distinguishes between performance evaluation and effectiveness evaluation by examining their core concepts, theoretical foundations, and international practices. Furthermore, it constructs a comprehensive evaluation framework encompassing multiple dimensions:smart spaces, smart services, smart resources, smart management, and smart librarians. Drawing on representative domestic and international case studies, this paper analyzes the key technologies, practical achievements, and challenges currently facing smart library development. It aims to provide theoretical references and practical pathways for the scientific evaluation and sustainable development of future smart libraries, ultimately driving the professional advancement of library services through management empowerment. This will enable libraries to demonstrate their irreplaceable social value in the era of digital intelligence.

Against the dual backdrop of policy impetus and the upgrading of public cultural demands, this study adopts the diffusion of innovations theory, resource-based view, and value co-creation theory as analytical frameworks to systematically examine the current state of cultural creative product development in Chinese public libraries, encompassing organizational structure, resource transformation, and marketing promotion. It delves into the core challenges faced, including weak IP development, a shortage of professional talent, limited marketing channels, insufficient funding, and imperfect copyright mechanisms. Drawing on case studies from both domestic and international contexts, this paper proposes systematic pathways for breakthrough from the perspectives of resource integration, brand building, technological innovation, and cross-sector collaboration, aiming to provide theoretical insights and practical references for transforming libraries from traditional document service institutions into modern engines of cultural innovation.

Digital resources are an indispensable strategic resource for technological innovation. In the context of insufficient funds, the digital resource guarantee of university libraries urgently needs to adapt to the changes in new information technology and user demands. Starting from the current situation, the main body, and subsystems, an in-depth analysis is conducted on the influencing factors and relationships of the digital resource guarantee mechanism of university libraries.The system dynamics model is used to describe the mechanism of each subsystem, Vensim PLE is utilized to simulate its dynamic changes, and a collaborative guarantee framework is constructed. The factors influencing the digital resource guarantee mechanism of university libraries can be divided into the obvious layer and the potential layer. It is proposed to help university libraries overcome the financial difficulties and enhance the sustainability of digital resource guarantee by reconfiguring the new pattern, reconstructing the new ecosystem, and reshaping the new business model.

This paper balances economic benefits and technical efficiency, accumulates domain data and artificial intelligence application experience, and explores a low-consumption, high-intelligence, and highly universal AI consulting construction plan. It analyzes the application status of online consulting platforms in domestic university libraries, proposes a construction idea of “low initial investment-meeting basic needs-accumulating domain data-intelligent upgrading in the later stage”, and explores a “human-intelligence collaboration” consulting path in two steps. In the first stage, an artificial + robot Q&A consulting platform is completed with low cost and high efficiency; in the second stage, the data volume and model structure are upgraded to build a “human-intelligence collaboration” consulting platform; this paper puts forward a three-stage collaborative knowledge service model to balance the collaborative efficiency between librarians and AI.

Given the discrepancies in research on influencing factors of existing scientific research collaboration intentions,conducting a comprehensive analysis of related empirical studies in this field holds significant importance for accelerating the development of scientific research collaboration. This article employs meta-analysis methods based on a random-effects model to examine 33 domestic and international empirical studies, identifying key factors influencing scientific research collaboration intentions, categorized into personal and environmental factors, as well as analyzing the relationships between these factors and scientific research collaboration intentions. Through main effect analysis and moderation analysis, it is found that critical personal factors affecting scientific research collaboration intentions include self-efficacy, perceived usefulness, and perceived control; critical environmental factors encompass reciprocity and collaborative attitudes. Furthermore,the research environment significantly moderates scientists’ collaborative intentions. Finally, based on data analysis, this paper proposes strategies for promoting scientific research collaboration.