Purpose/significance Establishing a specialized and unified research object for China’s information science can resolve the differences arising from “what is information” and the resulting problems of complex or unclear research objects，and foster the integration of various information research fields. Method/process This paper first sorts out and analyzes the evolution process and the internal reasons for the variability of the research objects in information science，and then expounds the theoretical basis for the unification of numerous research objects in information science into facts. Result/conclusion Situations，documents，materials，knowledge，information，data and other elements have successively become the research objects of information science. They are unified in facts，thus facts are the research object of information science. Establishing facts as the research object of information science helps China’s information science move beyond the debates surrounding the two “I”s （Information and Intelligence） and the two “IS”s （Information Science and Intelligence Studies），and advance the development of Chinese information science.

Purpose/significance Integrating signal strength for technology opportunity discovery helps to overcome the limitations of traditional technology opportunity identification，which focuses predominantly on explicit features. By leveraging the diverse characteristics of heterogeneous networks，this approach enriches the methods for discovering technology opportunities and provides valuable references for academic research and the formulation of science and technology policies. Method/process Firstly，after measuring the signal strength of technological themes in the application domain based on their novelty，value，sustainability，and attention level，a time-aware heterogeneous network is constructed comprising three types of nodes：patent classification codes，patents，and topics. Different meta-paths are designed to capture the multi-dimensional associations between nodes. Subsequently，nodes are initialized with embeddings by integrating content features and semantic features. Finally，the feature representations of patent classification codes are learned using meta-paths and node attributes，and technology opportunities are discovered through calculating vector similarities. Result/conclusion Taking the artificial intelligence field as an example for empirical analysis，the results show that the method of technology opportunity discovery by integrating signal strength in heterogeneous networks can effectively reveal the development direction of technical themes within the field，identify potential future technology opportunities，and provide strong support for scientific research，thereby promoting technological innovation and breakthroughs.

Purpose/significance To address the analytical challenges posed by massive unstructured military exercise intelligence and enhance the automated monitoring and early warning capabilities for national security risks，this study focuses on researching threat identification methods for military exercises. Method/process A hybrid entity extraction approach combining rule-based and deep learning methods is adopted to build a structured military exercise corpus. On this basis，a feature system covering eight dimensions is constructed，and the XGBoost model is employed for threat identification. The proposed framework is systematically validated through comparison with benchmark models，dimensional ablation experiments，and feature importance analysis. Result/conclusion Experimental results show that the XGBoost model performs well in threat identification tasks. The study confirms that，compared with the officially stated exercise purposes，such elements of the exercise as the scale of forces and exercise subjects，namely its physical capabilities and actual actions，serve as more reliable bases for threat identification. This conclusion provides a new data-driven analytical perspective for assessing the true intentions and potential risks of military exercises. Innovation/limitation This research innovatively constructs a hybrid entity extraction framework and establishes a threat identification system for military exercises. Current limitations lie in the reliance on open-source textual data and the static nature of the evaluation. Future work may extend toward multimodal data fusion and dynamic analysis.

Purpose/significance This paper explores the influencing factors of the collaborative innovation performance of the national key laboratory and analyzes the causal coupling paths. It aims to provide theoretical basis and practical references for the optimization of resource allocation，the improvement of collaborative governance mechanisms，and the enhancement of collaborative innovation efficiency of the national key laboratory. Method/process Through literature review，this paper systematically constructs an impact factor system of 19 indicators，covering four dimensions：innovation investment，organizational management，collaborative governance，and external environment. Subsequently，the DEMATEL method is employed to identify the core driving factors，and in combination with ISM，a multi-level structural model is constructed. From the deep，middle，and surface levels，the logic of the impact factors’ effects is revealed，and based on this，the key transmission paths influencing the collaborative innovation performance are refined. Result/conclusion The regional economic development level，the government’s policy support intensity，and the scale of research funds are the deep driving forces for collaborative innovation； the management of research plans，the autonomy of core equipment，the sharing of research data，the proportion of research personnel，and the performance evaluation orientation constitute the core mid-level support for collaborative innovation. Team collaboration efficiency，the stability of cooperation entities，and the number of innovation service institutions present the surface result characteristics，revealing the overall chain characteristics of macro-environmental driving，mid-level governance undertaking，and micro-level collaboration implementation. Innovation/limitation This paper introduces the DEMATEL-ISM method to systematically identify and hierarchically analyze the key influencing factors of the collaborative innovation performance of national key laboratories，and reveal the interaction paths and structural mechanisms. Expert ratings are subjective and the sample size is limited. The model is based on cross-sectional data and fails to reflect the dynamic changes in factor relationships. Although the influencing factors have been systematically summarized，some situational or latent variables may still be omitted，and the extrapolation of the results requires further verification.

Purpose/significance The construction of a digital government is a core component of the digital China strategy. Research on the characteristics of its spatial correlation network helps reveal collaborative relationships between regions and provides a scientific basis for optimizing the overall layout of digital government construction. Method/process From a systems theory perspective，this study constructs an evaluation index system for the development level of digital government construction. Integrating complex network theory，it employs a modified gravity model to build a spatial correlation network for digital government development using panel data from 31 provinces across China from 2017 to 2024. Utilizing social network analysis methods，the study examines network characteristics through overall network analysis，individual network analysis，and block model analysis. It explores the spatial correlation relationships，core nodes，and spatial spillover effects in digital government development across different regions. Result/conclusion China’s digital government development exhibits a networked evolutionary trend characterized by enhanced overall connectivity and improved regional coordination. The spatial network remains relatively sparse，and network efficiency still has room for improvement. Eastern coastal regions have formed a stable core network structure with strong capacity for information diffusion and resource integration. The intermediary role of central and western regions has gradually strengthened，while some peripheral provinces show structurally weak connections. Block model analysis further reveals a multi-level spatial pattern involving net spillovers，bidirectional spillovers，and broker roles，and finds that several provinces have achieved role transitions，reflecting the spatial restructuring of regional digital governance momentum and the continuous optimization of collaborative mechanisms.

Purpose/significance Combining theoretical modeling with generative AI applications to systematically analyze the future development trends and challenges of government data governance can help enhance the governance effectiveness of the government in complex digital environments. Method/process A three-stage research path of “theoretical modeling-technical empowerment-trend deduction” is proposed to achieve the synergy between theoretical analysis and generative AI. First，based on the “subject-time series” two-dimensional theoretical model，the evolution process of the government data governance structure is systematically modeled to reveal the dynamic evolution mechanism of the governance structure。Secondly，the GraphRAG technology，which has important application value in the field of generative AI，is introduced to construct a knowledge graph of government data governance，accurately depicting the temporal evolution characteristics of subject feature differences and driving factors，and providing empirical evidence for trend inference. Finally，taking the fusion of semantic elements as the starting point，cross-subject and cross-time series interaction paths and multiple evolution scenarios are explored，thereby realizing forward-looking governance trend deduction based on structural logic. Result/conclusion Based on a collaborative analysis of theory and technology，we reveal the overall pattern of the structural evolution of government data governance：an evolution from functional separation to structural integration in the subject dimension，and an upgrade from static rules to dynamic mechanisms in the temporal dimension. This further deduces the future development direction of government data governance in terms of platform collaboration，trust mechanisms，intelligent compliance，and public participation.

Purpose/significance This study explores the formation mechanisms behind graduate students’ dependence on generative AI in the context of its deep integration into academia，offering insights to enhance human-AI collaboration in academic innovation. Method/process Using grounded theory，we analyzed semi-structured interview data from 24 postgraduate students through three rounds of coding to identify key influencing factors and manifestations of dependence，and to construct a mechanism model. Result/conclusion Task constraints，contextual norms，and technological affordances serve as external factors that drive cognitive needs. Academic literacy plays a moderating role in the transformation of these needs，which ultimately leads to the formation of dependence on generative AI. This dependence manifests through behavioral path anchoring，cognitive inertia，psychological dysregulation，and academic social isolation. ［Innovation/ value Focusing on the phenomenon of generative AI dependence，this study systematically elucidates the mechanism of “environmental induction，literacy moderation，need-driven formation，and behavioral solidification.” It offers valuable insights for understanding the evolution of academic practices in human–AI collaboration and for developing targeted risk mitigation strategies.

Purpose/significance In the new era，intelligence resources serve as strategic assets for promoting innovation and safeguarding security in the field of science and technology. The construction of multimodal resources characterized by security，independence，and controllability is emerging as an urgent requirement for the implementation of strategies that coordinate the security and development. Method/process On the basis of a systematic review of intelligence resources，this study proposes the connotative characteristics of scientific and technological intelligence resources. By synthesizing the dual changes in the co-opetition environment and resource content，the primary security risks faced by resource construction in the new era are identified. The demand characteristics of resource development are clarified from three aspects： external demands，resource expansion，and technological driving forces. Result/conclusion Guided by the basic principle of coordinating security and development in science and technology，this study integrates intelligence thinking into the entire chain of resource management. Furthermore，feasible measures for resource construction are proposed，including the coordination of resource construction and security，the compilation of dynamic resource catalogues，the empowerment of digital intelligence technologies，and the assessment and verification of intelligence resources. The objective is to enhance national capabilities in the construction and security maintenance of intelligence resources for co-opetition in science and technology.

Purpose/significance In response to the construction needs of smart libraries during the “15th Five-Year Plan” period in China，this paper is committed to Introducing embodied intelligence technology to enhance users’ embodied experience of smart library services，thereby resolving the long-standing limitation of “disembodiment” in library services and driving the upgrading of smart library services. Method/process Focusing on the integrated development mechanism of embodied intelligence and smart library services，this paper discusses the theoretical logic of introducing embodied intelligence into smart libraries，and analyzes the application characteristics of embodied intelligence in smart library services. It constructs an convergent application framework of embodied intelligence and smart libraries from the “three-layer four-dimensional” perspective of “perception-decision-making-execution” and “resource-user-space-emotion”. Result/conclusion This paper explores the practical path of the transformation of library service paradigms driven by embodied intelligence，with the aim of providing useful references for the positive application of embodied intelligence in the development of smart libraries.

Purpose/significance The academic community acts as the core agent and anchor for governing the academic evaluation ecology in philosophy and social sciences. Method/process Drawing on an ecological perspective，and grounded in the construction of an autonomous knowledge system，we utilize the community’s diverse composition and normative principles as a baseline to identify the multiple challenges in evaluation ecology governance and propose corresponding solutions. Result/conclusion In this ecology，energy flow relates to academic resource allocation，material cycling focuses on knowledge innovation，information transfer relates to the circulation of evaluation data （standards，processes，and results），and balanced regulation centers on system stability. The core problem is the community’s dysfunction，specifically manifested as：energy flow deviation due to a shift in value orientation； material cycling imbalance resulting from the solidification of the evaluation landscape； disorder in information transfer caused by interest interference； failure of balanced regulation due to external interventions （e.g.，administrative control and market penetration）. To address this，we propose four governance pathways：prioritizing intellectual autonomy and originality in evaluation； fostering pluralistic co-governance to drive innovation； establishing a just and trustworthy evaluation relationship order； and cultivating a self-aware and self-consistent academic field.

Purpose/significance In the era of digital intelligence characterized by massive，high-dimensional，and real-time dynamic information，traditional information adoption theories are facing challenges to their applicability. This study focuses on the Data Storytelling （DST） methodology with the aim of revealing the underlying mechanism that drives users’ information adoption behavior，thereby constructing an information adoption theoretical framework adapted to the new environment and providing theoretical support for optimizing information dissemination effectiveness. Method/process First，three core elements of DST—data analysis，storytelling modeling，and narrative presentation—were clearly identified. These elements were then respectively mapped onto the central route，peripheral route，and mediating variables in the Information Adoption Model （IAM），resulting in the construction of DST-based theoretical model of information adoption behavior. Furthermore，two distinctive psychological pathways of information adoption shaped by DST—emotional resonance and cognitive change—were proposed and elaborated. Result/conclusion An empirical analysis was conducted using the Nature150 case to verify the proposed information adoption framework. The findings indicate that the three core elements of DST can jointly promote users’ information adoption behavior through the dual psychological pathways of emotional resonance and cognitive change.

Purpose/significance Based on a systematic analysis of the Regulation on a European Health Data Space （EHDS），this study refines its key practices in the cross-border flow of health data to provide reference for the construction of a trusted cross-border health data space in China. Method/process By deconstructing the content of the policy text，the data element value conversion process of primary and support activities in the EHDS is sorted out using the data value chain theory. A value co-creation perspective is introduced for supplementary analysis in links such as data collection，organization，circulation，and utilization. Result/conclusion Based on the construction practice of China’s trusted cross-border health data space and drawing on relevant experience from the EHDS，this study proposes the establishment of a unified data collection specification，establishing a full-process traceable quality assurance framework relying on data organizations，designing a secure and controllable health data circulation plan，and constructing a scenario-driven efficiency-enhancing data utilization system. Innovation/value Focusing on the specific links of data element value conversion in the EHDS，this study enriches the theoretical research perspective of trusted data spaces while realizing the trusted circulation and efficient utilization of cross-border health data.

Purpose/significance Large Language Models （LLMs） are extending the cognitive boundaries of intelligence decision-making systems through intelligent transformations in knowledge production mechanisms and human-machine interaction paradigms. This study investigates the paradigmatic shift in intelligence decision-making enabled by LLMs and its associated cognitive bias challenges. Method/process First，we systematically trace the technological evolution of LLMs and elucidate their three-tiered transformative impacts on intelligence decision-making：breakthroughs in data processing capabilities at the technical level，innovation in decision-making paradigms at the theoretical level，and the emergence of human-machine collaborative wisdom generation at the cognitive level. Second，we conduct an in-depth analysis of the mechanisms underlying cognitive biases in LLMs-assisted decision-making，constructing a classification taxonomy encompassing four novel bias categories：technology-induced biases，technology-amplified biases，human-machine collaborative biases，and data-derived biases. We further reveal how these biases disrupt the Data-Information-Knowledge-Wisdom （DIKW） chain through mechanisms such as cognitive path reconstruction，intelligent amplification effects，responsibility vacuums zones，and data “original sin” transmission. Result/conclusion We propose a cognitive tension equilibrium framework integrating dual process theory and cognitive ergonomics，employing strategies such as transparency labeling，multi-path verification，responsibility boundary demarcation，and dynamic data repair to address the fracture points induced by these biases. This research innovatively establishes a cognitive bias taxonomy for LLMs-driven intelligence decision-making scenarios and embeds dual-process theory and cognitive ergonomics into LLMs-enabled decision-making research，introducing a dynamic equilibrium intervention strategy based on “cognitive tension”. Our findings not only provide a novel analytical perspective for understanding cognitive phenomena in complex human-machine interactions but also offer cognitive ergonomic guidelines for designing next-generation intelligent intelligence analysis and decision-making platforms.

Purpose/significance As a strategic emerging industry，the low-altitude economy is characterized by strong technological convergence，rapid evolution，and complex integration pathways. Static analyses based solely on existing patent landscapes are insufficient to identify potential future technology combinations. Therefore，this study proposes a technology theme identification framework integrating link prediction and community detection to enable forward-looking identification of future technology themes and potential convergence paths in the low-altitude economy. Method/process Based on global low-altitude economy patent data （2004–2024），historical IPC co-occurrence networks are established using International Patent Classification （IPC） codes as semantic units. Graph Neural Networks （GNN） are then employed for link prediction to identify high-probability latent technological associations absent in current networks. The 2025–2027 predictive technology networks are subsequently generated. Louvain algorithm is applied for community detection on both historical and predictive networks，and future technology themes are identified by comparing the evolution of community structures. Result/Conclusion The historical technology system of the low-altitude economy remains primarily centered on breakthroughs in single-platform capabilities，including aircraft control，unconventional configurations and vertical take-off and landing layouts，and propulsion principles. In contrast，the predicted network reveals significantly enhanced coupling among technology communities，indicating a shift of industrial focus from equipment development toward system-level operational capabilities. Four major future technology themes are identified： integrated coordination of flight control，propulsion，perception，and mission execution； multi-domain operation architectures； intelligent airspace management and scheduling； and full life-cycle operational support systems. Innovation/value This study proposes a three-stage analytical framework of technology network construction–development trend prediction–key theme identification，integrating IPC co-occurrence networks，graph neural network-based link prediction，and community detection methods. The framework enables systematic modeling from technological relationship prediction to technology theme identification，providing data-driven support for technology planning and industrial strategy in the low-altitude economy，and offering a transferable methodological approach for forward-looking technology identification in other strategic emerging industries.

Purpose/significance Interactive short texts in digital communities can reveal group interactions，social dynamics，and public needs. However，existing studies remain limited to macro-level trend analysis and attitude profiling，lacking systematic recognition of multi-event coexistence，fine-grained semantic features，and effective visualization of cross-topic dynamic evolution. Method/process This study proposes a topic identification and visualization method tailored for interactive short texts. First，an initial event detector is constructed，and core events are identified through logical matching and contribution analysis of short texts to events. Second，by integrating speaker and textual features，a self-attention convolutional neural network is designed to learn the relationships between sentence pairs within a dynamic window，thereby achieving short text–event matching. Finally，event node information is extracted from event sets using the Qwen3 large language model，and a dynamic event evolution graph is constructed to efficiently present topic evolution. Result/conclusion Experiments on WeChat community group chat datasets demonstrate that the proposed method effectively visualizes topic evolution，offering valuable references for providing precise and proactive community governance services. Innovation/value This study introduces dynamic sentence-pair modeling to alleviate semantic sparsity and the coexistence of multiple events in interactive short texts，and integrates event evolution graphs to structurally represent event causality and evolutionary relationships，thereby providing a novel methodological framework for short-text analysis and visualization-based decision support in community governance.

Purpose/significance The information cocoon，as a potential risk within the network information ecosystem，presents challenges for quantification due to its metaphorical characteristic of “being present yet intangible”. This paper aims to establish a collaborative criterion framework integrating semantic and topological spaces to achieve dynamic identification of information cocoon formation in social media，thereby providing theoretical foundations and methodological support for the governance of the network information ecosystem. Method/process The study first employs the Louvain algorithm to identify communities within users’ social complex networks，revealing group homophily from the topological space. Second，it uses LangChain to compute semantic distances between communities and combines Word2Vec and SKEP models for topic identification and sentiment analysis，uncovering content homophily from the semantic space. Finally，by calculating global clustering coefficients and motif-based topological metrics of the community complex network，it analyzes the evolutionary patterns of network structure，exposing selection homophily. Through these three synergistic criteria，the study comprehensively evaluates the formation dynamics of information cocoons. Result/conclusion The semantic topological space collaborative criterion method developed through research has effectively identified the information cocoon situation in social media. In empirical research，identifying different forms of cocoons with different risks not only recognizes the existence of cocoons，but also reveals the strengthening process of their internal situation. Innovation/value The theoretical innovation of this article lies in proposing a criterion framework for “semantic topological dual space collaboration”，which transforms the abstract concept of information cocoons into three homogenized computable indicators of “group content selection”，providing a new path to solve its quantification problem.

Purpose/significance This paper proposes a new entity relationship extraction method based on large language models（LLMs） to address issues such as domain adaptation and data scarcity in entity relationship extraction in vertical fields. Method/process We design the self-optimizing chain of thought （SO-CoT） generation mechanism，introduce the multi-model collaborative fact evaluation strategy，and conduct comprehensive evaluation experiments using three mainstream LLMs on both a self-built Chinese ship faults corpus and the publicly available CMeIE-V2 dataset. Result/conclusion On the self-built corpus，the F1 score of DeepSeek-R1 is 88.35%，which is superior to other LLMs and mainstream deep learning. Following optimization using the proposed method，the F1 scores of ERNIE-4.0，GPT-4 and Deepseek-R1 are increased by 10.98，12.22 and 13.11 percentage points over their respective baselines. On the public dataset，the F1 score of the DeepSeek-R1 has increased by 15.89 percentage points over the baseline. The results show that the SO-CoT prompt outperforms conventional handcrafted prompts and exhibits notable adaptability. The fact evaluation mechanism provides a measurable performance gain. Our method significantly enhances the performance of LLMs while maintaining strong robustness. Innovation/value This work presents a universally applicable and efficient framework for entity relationship extraction in low-resource vertical domains，leveraging prompt self-optimization and multi-dimensional evaluation to enhance the performance of large language models.

Purpose/significance This paper proposes a large language model-based Agents framework for the task of academic text knowledge mining，seeking to address limitations of current large language models in this task，including insufficient task adaptation，outdated information and knowledge gaps，and a lack of planning and reasoning. By improving the performance of large language models on academic text knowledge mining tasks，the proposed method can provide theoretical and tool support for expanding and enriching the research methodological system of academic text knowledge mining under the paradigm of large language model. Method/process First，the paper introduces AcadMineGPT，a domain-specific large language model （LLM） for the task of knowledge mining from academic texts，improving accuracy and task adaptability of the large language model in this task. Second，a multi-agent system driven by this domain-specific LLM is proposed to automate and optimize the complex academic text knowledge mining process，further improves its real-time information acquisition ability in this task and the planning and reasoning ability to handle complex tasks. Finally，an evaluation framework for the tasks of academic text knowledge mining is constructed，which aims to systematically and comprehensively evaluate the performance of the methods proposed in this paper，and compares and analyzes it with mainstream LLMs. Result/conclusion Experimental results show that compared with the baseline models，the method proposed in this paper effectively improves the performance of academic text knowledge mining tasks and achieves more professional academic text knowledge discovery.

Purpose/significance The rapid development of large language model technology has profoundly impacted scientific and technical intelligence work. To construct a knowledge base for high-quality and diversified instruction data are crucial for enhancing the professional instruction-following capabilities of large language models. Method/process Based on the 5W1H analysis method，the critical issues concerning the construction of an instruction knowledge base in improving the training and application of large language models tailored for scientific and technical intelligence work are investigated. Result/conclusion The construction of an instruction knowledge base for scientific and technical intelligence work is not only necessary but also theoretically and practically feasible. Based on the specificity of instruction data in the field of scientific and technological intelligence work，and propose an knowledge organization solution for instruction data to provide guidance for knowledge base construction.

Purpose/significance In the context of the digital economy，the massive and complex volume of user-generated data poses challenges for traditional user profiling methods in terms of deep feature extraction and semantic understanding，making it difficult to support fine-grained demand mining and precision marketing. Method/process This study proposes a user profile feature extraction method based on multi-agent collaborative reasoning. Low-rank adaptation is used to perform multi-task instruction fine-tuning on a general large language model； a progressive collaboration chain centred on “experience–emotion–personality” is designed，and retrieval-augmented generation，short-term memory and quality-check mechanisms are incorporated to enhance robustness and structured outputs. Comparative and ablation experiments are conducted on approximately 15000 comments from 15 anime titles against multiple large language models and their fine-tuned variants. Result/conclusion Our method outperform baseline models in terms of accuracy and robustness across all three tasks，with the collaboration order experience–emotion–personality“experience–emotion–personality”. Innovation/value The study depicts users’ deep characteristics from an integrated three-dimensional perspective of experiential perception，emotional state and personality traits，and combines instruction-tuned large language models with multi-agent collaborative reasoning，providing a reusable technical pathway for demand mining and personalised recommendation in complex review-text scenarios.

Purpose/significance This study introduces a framework for examining technology convergence trends by dynamic example prompting and LoRA fine-tuning. By accurately pinpointing these trends，the paper aims to offer valuable insights for industrial policy formulation and corporate innovation strategies. Method/process The proposed method refines a large language model using a dual fine-tuning approach，incorporating dynamic examples prompting and LoRA technology. This enhances the extraction of technical terms. By amalgamating co-occurrence and semantic features of these terms，a dynamic fusion network is constructed. The Louvain algorithm is then employed to discern technology communities. Subsequently，indices measuring fusion intensity and diversity are devised to evaluate community convergence levels. Drawing from technology development and exploration theory，these communities are categorized into four fusion modes： growth，decline，exploration，and emergence. These categories aid in understanding community convergence trajectories. Result/conclusion When implemented in the domain of intelligent connected vehicles，the optimized Llama3.1-8B model，via the dual fine-tuning strategy，outperforms its counterparts with an F1 score of 71.9%—a marked improvement of 20.3% over the baseline model. The depth and breadth of technology convergence within this field have shown consistent enhancement. Furthermore，the convergence process has evolved from mere module linkage to intricate system coupling. Notably，there has been a shift from a predominantly growth-oriented mode to one characterized by exploration and emergence. Analysis of the technology convergence emergence mode from 2022-2024 reveals three primary fusion directions for future advancements in this sector： digital twin-driven safety assessment，the deepening of a multi-source integrated testing and verification system，and the integration of data security with collaborative control.

Purpose/significance Accurate measurement of innovation in academic papers is crucial for optimizing research evaluation and incentivizing breakthrough research. This paper aims to conduct a systematic review of domestic and international quantitative measurement methods for innovation in academic papers，thereby laying a solid foundation for the continuous innovation and refinement of future research and measurement methods. Method/process Based on the “data-driven” characteristic of academic paper evaluation research，existing measurement methods are categorized into two types：those based on full-text corpora and those based on citation corpora. This paper deeply reviews the developmental trajectory，scope of application，and limitations of these two categories，and summarizes their applications in comprehensive evaluation. Result/conclusion Currently，both categories of methods have achieved research progress. The former，focusing on the “novelty” dimension of innovativeness，is relatively well-researched and mature，having evolved from term frequency statistics to the mining and analysis of deep semantic representations and network structures. The latter，focusing on the “usefulness” dimension，is relatively lagging due to the difficulty in acquiring citation corpora，leaving significant room for improvement. These two types of methods are suitable for immediate and retrospective evaluation tasks，respectively. Future breakthroughs should be sought in three aspects：first，further clarifying the conceptual connotation of “innovativeness” to strengthen the theoretical basis of measurement research； second，vigorously addressing the deficiencies in “usefulness” measurement； and third，actively exploring the application of Large Language Models （LLMs） and AIGC tools to effectively empower the optimization and refinement of measurement methods.