[Purpose/Significance] To address the issues of insufficient datasets and difficulty in cross-domain transfer in Chinese aspect-based sentiment analysis (ABSA), the paper explores the application and performance of Chinese large language models in ABSA tasks. [Method/Process] The paper utilized the ChatGLM model, employing prompt engineering and fine-tuning techniques of LoRa and P-Tuning, to conduct ABSA tasks on the Chinese open aspect-level sentiment analysis dataset, ASAP. [Result/Conclusion] Compared with baseline models, ChatGLM based on few-shot prompts has performance close to that of full-sample deep learning models, while models combined with LoRa and P-Tuning achieve the best results and show good classification ability in the actual ASAP dataset.The paper verifies the effectiveness and feasibility of aspect-based sentiment analysis of Chinese large language models in the Chinese context and provides new solutions and references for the field of Chinese ABSA.
[Purpose/Significance] Patent inventors often face information overload when seeking collaborators, and collaboration among patent inventors is frequently constrained by various practical factors.Identifying potential collaborators for patent inventors amidst vast amounts of information is a pressing issue that needs to be addressed at this stage. [Method/Process] This study constructed a collaboration network for patent inventors based on three layers of relationships: direct trust, indirect trust, and domain preferences.By integrating the content information of patent texts, a heterogeneous graph convolutional network was employed to identify potential collaborators for patent inventors, followed by a segmentation of these potential partners. [Result/Conclusion] Empirical analysis using patent data from the intelligent sensor field within the industrial internet demonstrates that the proposed method has strong practical application.The model's AUC, accuracy, precision, and recall rates significantly outperform baseline models.By identifying and segmenting potential collaborators for patent inventors, this approach provides targeted recommendations for collaboration, which facilitates resource sharing, complementary strengths, and enhances innovation efficiency.
[Purpose/Significance] This study investigates the application effectiveness and technical potential of generative AI in Chinese bibliographic work, providing valuable insights for the technological empowerment of bibliography and the intelligent development of bibliographic practices. [Method/Process] Using the dual-model experimental approach with ERNIE Bot and DeepSeek, six typical bibliographic tasks were designed across three dimensions: document organization, document representation and document dissemination.These tasks were executed by generative AI models, with experimental results evaluated and analyzed through a multidimensional hybrid assessment framework. [Result/Conclusion] Generative AI demonstrates strong capabilities in Chinese bibliographic tasks, exhibiting higher technical adaptability in modern bibliography, structured tasks, and public-oriented scenarios.Future efforts should focus on constructing intelligent bibliographic systems with hybrid model architectures, overcoming capability bottlenecks in comprehensive bibliographic tasks and academic literature scenarios, while advancing theoretical and practical innovations in bibliographic control, recommended reading lists, and reading promotion.
[Purpose/Significance] To fully explore the semantic association between medical images and text, this paper employs an augmented heterogeneous graph fusion method to improve image feature representation effect and interactive image and text information to achieve feature fusion, thereby enhancing the performance of multimodal medical named entity recognition. [Method/Process] Firstly, RoBERTa and ResNet extracted features from medical text and images, respectively.Subsequently, a visual augment module extracted key information from images and filtered out irrelevant noise.After extracting the features, a heterogeneous graph was constructed by using text and image nodes along with their corresponding edges to capture fine-grained semantic associations between the two modalities.The fusion of medical multimodal features occurred through a self-attention mechanism, a cross-modal gating mechanism, and a position-wise feed-forward network.Finally, the experiment validated the effectiveness of entity recognition on a Chinese multimodal medical dataset. [Result/Conclusion] The RMGFM model constructed in this study achieves an F1 value of 88.99% on the Chinese multimodal medical dataset, which is an improvement of 5.52%, 5.28%, and 5.08% compared to the F1 values of the multimodal baseline models of UMT, AGBAN, and UMGF, respectively.Experiments show that the Ro-UMGF*+Manifold (RMGFM)model proposed in this study can effectively mine semantic associations between medical images and texts, and performs well in the task of recognizing multimodal medical entities in Chinese.
[Purpose/Significance] Analyzing the continuance use intention of generative AI college student users can provide theoretical references for optimizing the generative AI user experience and enhancing the new quality productive forces of the society. [Method/Processes] The article constructed the influencing factors model of generative AI college student users' continuance use, which was based on expectation confirmation theory and use and gratification theory, and took college students from five colleges and universities in Hangzhou as the research object, collected data through the questionnaire survey method, and verified the variables and hypotheses in the model by using SPSS 27.0 and AMOS 27.0. [Result/Conclusion] The study shows that: satisfaction is a key influencing factor on college student users' continuance use intention, and the factors of information quality, system quality, pleasure satisfaction, utilitarian satisfaction, and other factors have a significant direct positive effect on college student users' continuance use intention, and the expectation of confirmation, information quality, system quality and other factors can have a significant indirect positive effect on college student users' continuance use intention through satisfaction.
[Purpose/Significance] This research endeavors to explore the application of warning labels in the governance of false short videos, and to provide a basis for the improvement of the effect of warning labels in the governance of false short videos, as well as the governance of the content ecology of short video platforms. [Method/Processes] Based on SOR theory, using short videos as the research context, taking health information as an example and combining situational experiment (N=247), this paper aimed to explore how the forms (i.e., label position, label modality) ×contents (i.e., warning intensity, warning source) of warning labels affect the warning effects (i.e., users' perceived accuracy, perceived credibility), and then affected users' willingness to share. [Result/Conclusions] Warning labels significantly affect users' perceived accuracy and credibility, which in turn significantly reduce their willingness to share and trigger "implied truth effect". Regarding the label position, warning labels located in the lower part of the video have the strongest warning effect. Regarding the label mode, warning labels mixed with graphics and text have the strongest warning effect. Regarding the warning strength, warning labels with"false content"have the strongest warning effect. Regarding the warning source, warning labels with official labels have the strongest warning effect. Confirmation bias significantly moderates the effect of warning label forms and contents on users' perceived accuracy and trust.
[Purpose/Significance] The rapid development of generative artificial intelligence (GAI)has brought about revolutionary changes in the field of information retrieval. This paper aims to analyze the learning-related search of users in the GAI environment, and to explore the influence of cognitive styles and task types. [Method/Process] This paper employed user experiments and data analysis methods, utilizing the classification of field-dependence and field-independence cognitive styles. Two types of search tasks, comprehension-oriented and analytical evaluation-oriented were designed to collect data on users' search behaviors and learning outcomes. The data were analyzed using analysis of variance (ANOVA) to test for significant differences. [Result/Conclusion] In the environment of GAI, cognitive styles influence the sources of users' search query construction, the duration of reading generated results, search satisfaction, and perceived knowledge gain. Meanwhile, task types affect the length of search queries, the time spent on constructing search queries, and perceived knowledge gain.
[Purpose/Significance] In emergency management, intelligence comes first. In the face of the complex and ever-changing internal and external environments, emergency intelligence services play an important role as the "eyes and ears, vanguards, and staff officers" in the government's emergency management. They are the fundamental guarantee for achieving intelligent emergency response. Based on the agile theory, conducting theoretical research and evaluation practices on the government's emergency intelligence service capabilities is conducive to improving the level of emergency intelligence services and promoting the integrated development of emergency management and intelligent intelligence. [Method/Process] This article started with the practical dilemmas of the government's emergency intelligence services in China. Based on the connotations and characteristics of "agility" in the agile theory, it explored the composition of the government's emergency intelligence service capabilities, constructed an evaluation index system, and finally put forward strategies for enhancing the government's emergency intelligence service capabilities. [Result/Conclusion] This article constructs a framework model of the agile government's emergency intelligence service capabilities, which is composed of five dimensions: intelligence resource guarantee, intelligence business process, digital and intelligent technology empowerment, emergency intelligence coordination, and emergency intelligence function. It also establishes an evaluation index system for emergency intelligence service capabilities, including 5 first-level indicators, 22 second-level indicators, and 77 third-level indicators. Focusing on the three aspects of the digital and intelligent upgrading of the government's emergency intelligence services, the construction of a new intelligence resource guarantee system, and the establishment of an emergency intelligence coordination network, it puts forward targeted strategies for improving the emergency intelligence service capabilities.
[Purpose/Significance] The generative artificial intelligence technology has triggered the copyright infringement risk during the generative AI data training stage, and infringement litigation cases around the world have emerged. It is of great significance to propose regulatory and compliance measures to deal with the copyright infringement risk of generative AI data training in China, to balance the interests of AI industry and traditional content creation industry, and promote the sustainable development of GenAI technology. [Method/Process] Through the method of literature, comparison and empirical analysis, this paper revealed the regulatory systems of developed countries (areas). On the basis of comparing and learning from the regulatory experience of developed countries, this paper gave the countermeasures in combination with our national conditions. [Result/Conclusion] The study found that China is at the rapid stage of generative AI development. Chinese government should introduce adaptable administrative regulatory principles. Meanwhile, China should incorporate a flexible "fair use" system that is applicable for GenAI data training into the current copyright legal system. At the same time, the industry can also develop a Creative Commons License scheme applicable to GenAI data training as a cost saving compliance measure.
[Purpose/Significance] AI policies have a profound impact on promoting knowledge flow and stimulating collaborative innovation in cities. This study employs agent-based simulation method to analyze the investment strategies of policy instrument mix, addressing the shortcoming of existing research that focuses primarily on qualitative analysis. It is of great significance for assisting AI policy formulation, facilitating the effective use of resources, and enhancing urban collaborative innovation. [Method/Process] This research was based on the perspective of AI policy instrument mix. Based on the automatic extraction of policy instruments, the rules were established for cooperation network connections, policy investment and benefit, and cooperation willingness updates. It simulated the impact of investments in three policy instruments—financial support, legal regulations, and application promotion—on urban collaborative innovation. The study explored the optimal policy instrument mix strategies and analyzed the dynamic changes in policy investment across different stages. [Result/Conclusion] The study finds that the combination of AI policy instruments promotes urban collaborative innovation, with a rapid increase in the early stage and a slower growth in the later stage. Under the optimal strategy, legal regulations dominate the policy investment in the early stage, followed by financial investment, while in the later stage, the focus is mainly on the application promotion. And the density and clustering coefficient of the urban collaborative innovation network gradually increase, continuously giving rise to secondary nodes in addition to the leading nodes.
[Purpose/Significance] Trusted data spaces provide critical support for data circulation and utilization.This paper aims to explore the construction principles and development paths of trusted data spaces that align with China's national conditions, providing development momentum for national integrated data element market. [Method/Process] This paper reviewed the research status and practical progress of data spaces through a literature review and case examples.It analyzed, from multiple dimensions, the methods and approaches for addressing challenges in data circulation and utilization through trusted data spaces.Based on these analyses, it systematically discussed construction principles and development paths of trusted data spaces. [Result/Conclusion] This paper proposes BASIC construction principles of trusted data spaces, including balance, accountability, safe, interoperability and controllable, and explores the development paths of trusted data spaces in China from the perspectives of strategic layout, technologies, standards and regulations, innovation ecosystem, operation and supervision, and international exchanges.
[Purpose/Significance] The age of digital intelligence has created new challenges while laying out new scenarios for the government data sharing in megacities, exploring the government data sharing system of megacities is beneficial for empowering the construction of megacities with government data. [Method/Process] Combined socio-technical system theory and technology enactment framework, this paper proposed an analytical framework, extracted the core elements, then described the overall appearance of the government data sharing system in megacities, explored its essential characteristics, analyzed its operation mode and tested the effectiveness of the system operation with Shanghai as a case.This paper put forward specific optimization strategies for government data sharing system in megacities. [Result/Conclusion] This paper puts forward a"technology-system"analysis framework, separates out four core elements of hard instrument technology, soft operation technology, static system design and dynamic system operation, reveales the essence of digital twin in this system and analyzes the inherent technological integration, technological empowerment, institutional inclusiveness and mechanism driven operation mode.Finally, this paper proposes an optimization strategy for government data sharing in megacities with clear core, goals, principles and measures.
[Purpose/Significance] The investigation of the measurement scheme of image-sensitive data of social network users is of great significance for the expansion of the scope of data privacy research and the promotion of the healthy development of social network platforms. [Design/Methodology] First, a social network privacy image database was created by combining privacy perception with objective standards.Next, using convolutional neural networks and information entropy theory, a quantitative model for image-sensitive data of social network users was developed.The image data publicly released by Sina Weibo users was then obtained.The dataset was labeled using a self-constructed privacy classification table, and supervised learning was performed.Finally, Sina Weibo users were sampled based on their age characteristics.The risk of privacy leakage was analyzed and categorized for early warning purposes. [Findings/Conclusion] The user-age factor serves to counteract the tendency for individuals to disclose personal information on social networks.The sensitivity of social network user image data is ranked from highest to lowest, with the following categories being considered: personal property information, personal network communication information, personal medical and health information, personal location information, and personal identification information.
[Purpose/Significance] Based on the perspective of science and technology fund co-funding, the study explores the current situation of regional co-innovation, and provides useful references for the construction of an efficient and reasonable regional innovation synergy mechanism. [Method/Process] Using the 31 provincial-level administrative regions in mainland China in Web of Science database as the research object, the study investigated the layout and influence effects of regional joint innovation through complex network analysis and stepwise regression method, etc. [Result/Conclusion] The results show that in general, the eastern region has the highest number of papers and the highest academic influence in co-funding; the level of regional co-innovation in each region has been increasing year by year, and the dependence on co-funding has been decreasing; geographic proximity and resource complementarity are the main driving forces for inter-regional co-innovation in the form of co-funding; the results of benchmark regression find that the number of papers and academic influence have a significant promotion effect on regional S&T innovation level, with the number of papers having a stable and long-lasting effect, and the academic influence and the novelty of achievements having a greater but weaker effect, with the latter having a certain lag in its promotion effect; the heterogeneity test finds that the number of papers and the academic influence have the greatest promotion effect on the S&T innovation level of the eastern and western regions; Progressive impact analysis suggests that regional co-innovation in the form of co-funding can be a potential driver of new quality productive forces.
