Intelligence Theories and Methods

645 Identifying Emerging Topics in Disciplines by Integrating Network Structural Features Hot! Yang Jinqing, Luo Man, Cheng Xiufeng, Xia Lixin, Ma Tingcan DOI: 10.3772/j.issn.1000-0135.2025.06.001 Identifying emerging topics in disciplines is an effective method in the advancement of technological innovation and tracking trends in disciplinary development. The emergence of emerging topics in disciplines is a complex process influenced by both scientific communication and self-organizing networks, whereby the network structures of emerging topics in different disciplines acquire unique properties. Based on the common characteristics of scientific communication, this study integrated the global and local network structural features of disciplinary topics. A standard experimental dataset was generated by random matching. A multi-indicator weighted fusion method and a machine learning classification method were then used to identify emerging topics in the disciplines. The results indicate that the multi-indicator weighted fusion method is effective in identifying top-ranked influential disciplinary topics. However, the P@270 high-influence topics accounted for only 60%, which is lower than the optimal performance of the random forest classification model of 64.14%. This indicates that the machine learning classification method has a greater advantage in fitting complex processes, whereas the multi-indicator weighted fusion method is more suitable for tasks focusing on the most influential topics. The results of machine learning interpretability analysis show that a higher citation frequency, network influence, number of published papers, and higher influence of authors on journals contribute to the identification of high-influence topics, whereas a higher degree of structural mutation has a negative effect on identifying emerging topics in disciplines. 2025 Vol. 44 (6): 645-659 [Abstract] ( 8 ) HTML (1 KB)  PDF (4054 KB)  ( 3 )

660 Research Technology Readiness Assessment Methods for Emerging Technology Directions Based on Applied Basic Research Hot! Wang Chun, Leng Fuhai DOI: 10.3772/j.issn.1000-0135.2025.06.002 This study investigated the intelligence methods of objective and reproducible assessment of technology readiness for emerging technology directions based on applied basic research, with the aim of improving the long-term situation where technology readiness is rated by experts and is difficult to replicate, as well as solving the problems of inaccurate evaluations caused by vague criteria of technology readiness level (TRL) ratings. This study first analyzed the adjustment and application status of the TRL scale and investigated the differences between expert and intelligence indicator evaluations. Second, in response to scholars’ encouragement of applications and adaptions of the present TRL scale, to strengthen industrial discussions, this article discusses research types, literature data types, source databases of TRL proxy indicators, and TRL ranges involved in emerging technology directions in detail based on applied basic research. Through an overall comparison of the TRL rating criteria from multiple industries and organizations, this article proposes a hypothesis for the three-stage ratings of technology readiness for emerging technology directions and presents an analysis framework of “Research Topic + Time Sequence.” Then, the consistency of the analysis results were validated through the comprehensive applications of time series clustering, topic modeling & main research-topic judgment, and semantic main path analysis. By verifying with expert consultations, this article ultimately demonstrates the feasibility of intelligence methods. This study conducted empirical research using 11833 papers on lithium iron phosphate battery technology since 1997 from the Web of Science Core Collection as an example, confirming that research in this field has reached the third stage of technology readiness for emerging technology directions. Research on the objective and reproducible intelligence methods of technology readiness assessment for emerging technology directions is a beneficial supplement to existing methods. This study has positive significance for understanding the development trajectory of technology, allocating scientific and technological strategic resources, formulating future industrial technology roadmaps, and promoting technological innovation and industrial development. 2025 Vol. 44 (6): 660-674 [Abstract] ( 9 ) HTML (1 KB)  PDF (2851 KB)  ( 3 )

675 Interdisciplinary Literature Identification Method Based on an Improved Deep Learning Model Hot! Feng Ling, Pan Yuntao DOI: 10.3772/j.issn.1000-0135.2025.06.003 Effectively identifying interdisciplinary literature not only helps to timely grasp the research trend of interdisciplinary research and track scientific research activities in interdisciplinary fields in real time, but also provides strong support for scientific research decision-making. This paper proposes an interdisciplinary literature identification method based on an improved deep learning model according to the semantic intersection. First, a training dataset for interdisciplinary literature identification is obtained through “text merging.” Then, an improved deep-learning-based text classification model is proposed and trained on the training set. Finally, based on the trained model, a new literature is determined whether it is interdisciplinary. This study conducts empirical research on “Dental Materials” and “Computational Biology” datasets. The results indicate that the proposed method is effective in interdisciplinary literature identification, and the area under the curve (AUC) values calculated on the two datasets— “Dental Materials” and “Computational Biology” —reach 0.741 and 0.966, respectively. Compared with traditional deep-learning-based text classification methods, the proposed method can train interdisciplinary literature recognition models based on existing non-interdisciplinary literature without relying on any prior knowledge of interdiscipline. Thus, when a new literature appears, the proposed method can accurately distinguish whether it is interdisciplinary, achieving real-time monitoring of cutting-edge interdisciplinary fields with development potential. Additionally, there is a significant improvement in the efficiency of identifying interdisciplinary literature as compared with the traditional methods. 2025 Vol. 44 (6): 675-687 [Abstract] ( 3 ) HTML (1 KB)  PDF (1765 KB)  ( 4 )

688 Toward a Technical Topic Popularity Evaluation Framework Based on the ELO Model Hot! Chen Hongkan, Liu Jinchang, Bu Yi DOI: 10.3772/j.issn.1000-0135.2025.06.004 Evaluating the popularity of technical topics is of great significance for decision-makers to understand market and technical development trends. However, extant indicators for evaluating popularity or recognizing weak signals still suffer from four serious issues: a lack of forward-looking perspectives, subjective and challenging adjustment of time interval thresholds, predetermined disciplinary frameworks and granularity, and difficulty in assisting intelligent decision-making with output results directly. To this end, this study introduces the concept of “expected popularity” based on the Elo rating system (ELO) model and constructs a new method for evaluating the popularity of technical topics. First, we theoretically discuss the feasibility of applying ELO methods to these tasks and take the recognition of the popularity of technical topics in the field of carbon fiber as an example to showcase its effectiveness. Compared to existing methods, the method proposed in this paper enriches the evaluation framework and provides decision-makers with more intelligence-level support. 2025 Vol. 44 (6): 688-701 [Abstract] ( 8 ) HTML (1 KB)  PDF (2185 KB)  ( 3 )

702 Theoretical Basis and Empirical Research of Citation Discontinuance Hot! Li Hao, Hou Jianhua, Zhang Yang DOI: 10.3772/j.issn.1000-0135.2025.06.005 Citation analysis is an important method for tracking and evaluating the spread and impact of scientific knowledge. However, existing research often focuses only on the diffusion process of citations and neglects the phenomenon of citation discontinuance, which refers to a situation in which a document is not cited again within a certain period after its first citation. Conceptualizing and quantitatively studying this phenomenon requires the systematic construction of a theoretical foundation and an analytical framework for citation discontinuance. By considering theories such as the diffusion of innovations, literature obsolescence, scientific paradigm shifts, and knowledge evolution, this paper provides theoretical explanations for the phenomenon of citation discontinuance, analyzes its connotations, and distinguishes between three types of citation trajectories: complete, transient, and intermittent citation discontinuance. Based on this, this study designed a quantitative identification method for citation discontinuance and conducted an empirical analysis. The proposed concept of citation discontinuance further reveals the durability and nonlinear characteristics of citation diffusion, aiming to provide new insights for modeling citation diffusion. This study also constructed an analytical framework for citation discontinuance to help scientific research institutions and researchers better understand the impact and timeliness of research outcomes, with the expectation of further enhancing the explanatory power of citation analysis methods and citation diffusion research in practical applications, such as literature lifecycle management and scientific evaluation. 2025 Vol. 44 (6): 702-719 [Abstract] ( 7 ) HTML (1 KB)  PDF (6186 KB)  ( 5 )

720 Research on Key Patent Identification for Solving the Technology Dilemma of “Neck Stuck” Hot! Zhu Jiahui, Zhou Xiao, Wang Bo, Ren Qiaoyang, Wang Dan DOI: 10.3772/j.issn.1000-0135.2025.06.006 With foreign restrictions on technology and products in the chip field, science and technology leading enterprises and institutions in China are working together to help break through the “Neck Stuck” technologies to achieve localization replacement. The key patents with breakthrough potential in “Neck Stuck” technology problems in China must be identified and targeted fine-grained solutions must be provided to help China achieve technology breakthrough and layout technology innovation strategy. On the basis of generating Chinese and foreign patent technology-effect texts with a large model, this study identifies the domestic key patents that can solve the technology problems of “Neck Stuck” by comparing the technology-effect texts of foreign core and domestic patents. The research innovation is mainly reflected in two aspects: (1) in the generation of technology-effect texts, the use of a large model as a text generation tool makes up for the mechanical extraction method and effectively reduces the degree of expert participation; (2) when identifying key patents, the deep correspondence of “problem-solution” is considered, and the Chinese and foreign patents are benchmarked from the underlying logic of technology rather than the surface of technical text, such that the identified Chinese key patents are more problem-oriented. Finally, this study considers the field of AI chips as an example to conduct an empirical research that verifies the feasibility of the proposed method and provides an effective way for China to break through the “Neck Stuck” technology dilemma. 2025 Vol. 44 (6): 720-735 [Abstract] ( 8 ) HTML (1 KB)  PDF (2834 KB)  ( 4 )

Intelligence Technology and Application

736 Named Entity Recognition of Ancient Books Based on MU Sequence Labeling Hot! Xu Qiankun, Wang Dongbo, Liu Yutong, Huang Shuiqing DOI: 10.3772/j.issn.1000-0135.2025.06.007 The named entity recognition task is an important basic step in many downstream tasks in natural language processing. Ancient books, as carriers of Chinese civilization, not only contain a rich cultural heritage, but they are also an important source of historical wisdom and enlightenment for the future. Improving entity recognition accuracy in ancient texts promotes the structuring of ancient texts and knowledge systematization, as well as the intelligent use and development of ancient resources. First, we selected the Twenty-Four Histories dataset refined by the group as the original dataset and used the GujiBERT_FAN pre-training model to fine-tune the Sequence Labeling, Sequence Labeling_CRF, and Span-level Prediction methods to capture entity boundaries and types more accurately. Subsequently, the entities in ancient texts were recognized and predicted. Second, this study developed methods of merging and majority voting mechanisms for integrating with the prediction dataset and creating a new dataset. Based on the recognized entity dataset, we created a new dataset using merging and majority voting methods in combination with the prediction results of the Named Entity Recognition model. Finally, the Sequence Labeling, Sequence Labeling_CRF, and Span-level Prediction methods were trained to determine whether the prediction results of the entities were incorrect, and the model was fine-tuned using hinted concepts. To validate the method proposed in this study, the effectiveness of the model was verified using evaluation metrics, which showed that the addition of merging methods resulted in a significant increase in the recall rate of entity recognition, and most voting methods improved the model’s F1 value. 2025 Vol. 44 (6): 736-747 [Abstract] ( 4 ) HTML (1 KB)  PDF (1221 KB)  ( 1 )

748 Patent-BARTKPG: A Contrastive Learning-Based Approach for Chinese Keyphrase Patent Generation Hot! Ran Congjing, Liu Xingshen, Wang Haowei, Liang Yulian, Wang Fuxin DOI: 10.3772/j.issn.1000-0135.2025.06.008 The traditional extraction methods used to generate keyphrases for patents are not sufficiently accurate. This is primarily manifested as excessive reliance on the literal content in the text, redundant information in the generated sequence of keyphrases, and inconsistency with the target keyphrases. To address these issues, this study combines the unique corpus characteristics of Chinese-patent texts to achieve a more accurate generation of keyphrases. A two-stage model is proposed for extracting, generating, and reordering keyphrases from patents. Additionally, a contrastive learning training strategy is introduced in both stages to further enhance the performance of the model. Finally, a Chinese-patent bidirectional auto-regressive transformer for keyphrase generation (BARTKPG), named Patent-BARTKPG, is constructed to accurately generate keyphrases for Chinese-patent texts. In preliminary studies, Patent-BARTKPG significantly outperformed other keyphrase extraction and generation models in generating high-quality keyphrases for the Chinese-patent dataset. 2025 Vol. 44 (6): 748-760 [Abstract] ( 6 ) HTML (1 KB)  PDF (3961 KB)  ( 2 )

761 Optimization of LLM's Generation Strategies Based on Rich Semantic Tokens Hot! Cheng Qikai, Shi Xiang, Yu Fengchang, Huang Shengzhi DOI: 10.3772/j.issn.1000-0135.2025.06.009 In recent years, general-purpose large language model (LLM) technologies have made significant progress. However, their application in information science still faces challenges, including low inference efficiency and insufficient task adaptability. To address these issues, this paper systematically analyzes the generation mechanism of LLMs and introduces the concept of “Rich Semantic Tokens,” which describes tokens or token sequences that LLMs tend to generate during the process, characterized by semantic aggregation, contextual dependence, or task relevance. Based on this concept, we propose a collaborative generation strategy between large and small models, driven by generation preferences. Through the mining of Rich Semantic Tokens, a copying mechanism, and a dynamic validation strategy, we enable collaboration between small and large models, promoting a shift from word-by-word generation to the simultaneous generation of multiple tokens, thus enhancing generation efficiency and task adaptability. This study evaluated the proposed generation optimization strategy across three dimensions: generation performance, generalizability, and generation efficiency. Experimental results demonstrate that this strategy outperforms traditional generation optimization methods in multiple domain-specific tasks, including law, medicine, and news encyclopedias. This study provides a new theoretical foundation and practical pathway for optimizing LLM generation, improving task adaptability, and constructing trustworthy and reliable LLMs. 2025 Vol. 44 (6): 761-782 [Abstract] ( 6 ) HTML (1 KB)  PDF (4934 KB)  ( 0 )

Intelligence Reviews and Comments

783 Representative Works in Science and Technology Evaluation: Concept, Connotation, and Characteristics Hot! Zhang Zhen, Xu Xiaoting, Cheng Ying DOI: 10.3772/j.issn.1000-0135.2025.06.010 A representative work system is an important aspect of the science and technology evaluation reform in China’s new era. Existing research lacks a comprehensive, clear, and scientific understanding of representative works. Data were collected from multiple sources, including semi-structured interview texts, domestic and foreign documents, and online posts. Grounded theory was employed to distill the characteristics of representative works (basic quality, academic value, relevance, compliance, impact, and academic competitiveness), and construct a conceptual model of the relationships among these characteristics, further refining the definitions of representative works. Specifically, representative works are defined as research outcomes that meet basic quality and compliance requirements; possess academic value, impact, and relevance; and demonstrate the academic competitiveness of researchers. The study indicates that researchers establish impact through basic quality, academic value, relevance, and compliance of their outcomes, demonstrating their academic competitiveness through these characteristics. This research deepens the understanding of representative works and has important theoretical and practical implications for researchers, peer reviewers, and research management institutions that implement representative work systems. 2025 Vol. 44 (6): 783-796 [Abstract] ( 14 ) HTML (1 KB)  PDF (1058 KB)  ( 10 )