The open source of DeepSeek has enabled the application of artificial intelligence generated content (AIGC) to enter a rapid development stage. However, it has also increasingly highlighted many problems. This study delves into the specific problems of AIGC, constructs a collaborative governance mechanism from three aspects: goal collaboration, process collaboration, and inter‐subject collaboration, and provides an implementation path for multi‐subject collaborative governance. The study found that in the DeepSeek era, AIGC technology faces problems such as user privacy leakage, insufficient content quality assessment, and intellectual property and ethical conflicts during its application. This paper emphasizes that establishing a collaborative governance mechanism is a key way to deal with these problems in the long term. The government, industry, platform, and users should participate together to strengthen industry supervision, improve the self‐discipline review mechanism, and enhance AI literacy education, so as to jointly promote the long‐term stability and healthy development of the artificial intelligence industry. This study is of great significance to ensuring information security and promoting the healthy development of the artificial intelligence industry.

ABSTRACTToxic content on social media has posed a significant threat to user experience and societal stability. In contrast to explicitly toxic content, implicitly toxic content, lacking overt toxic words, is challenging to identify through single‐text features. Therefore, this study proposes a multi‐feature fusion algorithm to recognize implicit toxic content. Firstly, we collect various features for each post, including text content, likes, comments, user information and other features. Subsequently, employing a multi‐head attention mechanism, we extract and fuse these features. Then, utilizing ensemble learning algorithms, we identify implicitly toxic content based on the fused features. Finally, we analyze the decision‐making process of the model using an interpretable algorithm and derive the most critical features and factors for identifying implicitly toxic content. The experimental results show that the proposed algorithm outperforms algorithms such as BERT, TextCNN, and XGBoost, demonstrating the advantage of multiple features over a single text feature in recognizing implicitly toxic content. In addition, this study provides insights into the decision‐making process of the model and provides more effective toxic content management strategies for social media platforms.

Government agencies have increasingly established their official accounts to disseminate information and publish rumourrefuting messages (RRMs) on social media platforms. However, little is known about what factors facilitate users to engage in RRMs posted by government accounts. To bridge this gap, our study borrows the lens of persuasion theory to frame a research model and unmask the precursors that foster social media users to engage in RRMs. By analysing RRMs published by 10 influential government official accounts spanning 9 years, a field study on Sina Weibo finds that the text length of an RRM is associated with a higher probability of liking, commenting on, and sharing the RRM, while the inclusion of links in RRMs is negatively linked to user engagement. The effect of the existence of photos and videos on user engagement in RRMs depends on different engaging behaviours. The inclusion of emojis in RRMs helps shorten users' psychological distance from the authorities, thereby facilitating user engagement behaviours. Using rhetorical questions is associated with a higher level of user engagement (including liking and sharing) in RRMs by increasing personal relevance. This study offers new insights into online rumour governance and practical suggestions for promoting government social media publicity.

To prevent other types of mental health problems from being misclassified as depression, as well as to remedy the problem of inadequate resources for mental health consultations. This study first analyzes the types of different causes of mental health problems, providing an important basis for better understanding the diversity and complexity of this field. Subsequently, a machine learning approach was used to predict the potential causes of different types of mental health problems. This research provides new perspectives and methods for early identification and personalized treatment of mental health problems. The experimental results show that depression accounts for only 16.9% of mental health problems. In the prediction of the causes of mental health problems, the SVM method performed best in predicting the causes of mental health problems, outperforming 5 machine learning methods and 3 deep learning methods. Through these studies, we hope to prevent other types of mental health problems from being misclassified as depression and to remedy the lack of resources for mental health counseling. This will help increase the success rate of early intervention and provide better mental health support for patients.

ABSTRACTThe proliferation of Artificial Intelligence Generated Content (AIGC) poses significant challenges to user experience and information accuracy, especially on search engine websites(Guo et al., 2023). The current solution is to identify AIGC by machine learning algorithms or publicly available AI detection tools, whereas, machine learning(Wang & Wang, 2022) algorithms degrade in accuracy as more data is available and tools such as GPTZero perform poorly in the task of AIGC detection on social media. In this paper, we propose an EPCNN model to identify AIGC on search engine websites, which maintains good performance in large‐scale samples. The ERNIE model integrates cross‐domain knowledge and improves language understanding and generalization. We use ERNIE to extract text features, then use a feature pyramid network to capture semantic information at different levels, and finally use an end‐to‐end structure to connect ERNIE and the feature pyramid network to construct the EPCNN. Experimental results show that our proposed algorithm has high accuracy and the ability to handle large‐scale data compared with machine learning algorithms and AI detection tools.

[Purpose/Significance]Health science knowledge is of great value for the improvement of public health literacy and controlling false health information on the Internet.[Method/Process]This paper took haodf. com as the context to select 50407 pieces of health science knowledge under the categories of cancer and skin diseases as samples. Then,the study used LDA topic model to mine the topics,quantity and presentation form of health science knowledge,and analyzed the preference and quality demand of community users.[Result/Conclusion]The topic of the health science knowledge in the online medical community revolves around specific diseases. Knowledge about disease treatment,examination and technical solutions,disease research progress and health care is available in large quantities,while the supply of knowledge such as medical consultation and health consultation,health science education,and pathological knowledge are not sufficient. At the same time,there is still a lot of room for improvement in terms of popular expression,diversified presentation and practical value. This paper theoretically enriches the relevant research on health science knowledge and health information demand,and provides a basis for online medical community managers and health professionals to optimize the supply of health science knowledge.

1 引言 大数据已成为国家基础性战略资源和数字经济时代的新引擎.随着技术与应用的成熟,大数据与人工智能的结合已成为新的趋势,大数据管理与应用领域将成为大数据与智能化的桥梁.

[目的/意义]用户接触辟谣信息,并不能等同于用户接受辟谣信息.研究疫情防控背景下影响用户接受政府辟谣信息意愿的主要因素和相关路径,有助于政府精准治理网络谣言.[方法/过程]调查疫情防控背景下的网络用户接受政府辟谣信息的相关心理与行为,以脆弱性理论为基础构建理论模型,运用模糊集定性比较分析法对问卷数据进行影响因素与组态路径分析.[结果/结论]政府辟谣信息接受意愿的影响因素组态可分为三类:社会支持型、社会支持-谣言鉴别能力型、谣言鉴别能力型.动态压力、信息焦虑、社会支持和谣言鉴别能力是影响政府辟谣信息接受意愿的核心因素.[创新/局限]组态分析更加深刻地揭示影响政府辟谣信息接受意愿的不同核心因素的联合影响机制和实现路径.后续研究可比较分析疫情不同时期影响用户辟谣信息接受意愿路径的差异性.

[Purpose/Significance] To explore the problems existing in the application process of algorithm recommendation service from the perspective of user’s behavior can provide guidance and norms for dealing with the chaos of algorithm recommendation. [Method/Process] The conditional value method was used to quantify the acceptance willingness of network users’ algorithmic recommendation service, and verify whether their acceptance intention deviates from their use behavior. Finally, the influencing factors of this phenomenon were analyzed based on the theory of perceived value and risk. [Result/Conclusion] In the process of using the algorithm recommendation service, there is a deviation between the willingness to accept and the use behavior of network users, that is, the user continues to use but the willingness to accept is not high. User perceived value has a negative effect on acceptance willingness and behavior deviation, and the user’s perceived pleasure has the most significant impact; perceived risk positively affects the deviation phenomenon, in which the degree of economic loss has the greatest impact, and the degree of information overload has the least impact.

[Purpose/significance] Exploring the mechanism of influencing factors of personal information sense of security will help to enrich the concept definition and theoretical framework of personal information security and other related fields, and effectively improve the level of citizens’ personal information sense of security.[Method/process] Comprehensively adopt Grounded theory, Qualitative Comparative Analysis and other methods to define the conceptual composition of personal information sense of security, build a theoretical framework of influencing factors of personal information sense of security, and explore the configuration mode of influencing factors.[Result/conclusion] It is found that personal information sense of security consists of risk experience, coping experience, solution experience, security experience and recall experience; the influencing factors theoretical framework of personal information sense of security covers three dimensions: macro environmental factors, processing subject factors and personal factors, including seven main category factors such as information ecology and personal information processor behavior; high personal information sense of security has two configuration modes: situational security mode and self-regulation mode.

[Purpose/significance]The public online health information search is increasing,and short video platforms have become an important health information medium.There exists a large amount of false health information on the short video platforms,which may be adopted by users.Therefore,analyzing driving factors of false health information adoption of short video users is conducive to the governance of false health information on the short video platforms and the optimization of online health information ecology.[Method/process]To clarify the driving factors of false health in-formation adoption behavior of short video users,this study has contextualized opportunity appraisals(perceived emo-tional support and perceived information superiority),threat appraisals(perceived risk and perceived barriers)and sec-ondary appraisals(health information literacy)under the short video platform context based on coping theory.[Result/conclusion]For primary appraisal,in opportunity appraisal,short video users'perceived information superiority exerts positive significant effect on their false health information adoption,while in threat appraisal,short video users'per-ceived risk exerts negative significant effect on their false health information adoption.For secondary appraisal,the health information literacy of short video users negatively and significantly affects their false health information adop-tion.This study has enriched the online community health information research,extended the research context of coping theory,and clarified the driving factors of false health information adoption of short video users.

[目的/意义]算法社会的来临对个人发展提出了新的目标要求,而算法素养则有助于增强个人与算法交互的能力,为个体提升算法认知和技能、应对算法社会风险、强化算法源头治理开辟了新的路径.鉴于现有研究中的概念局限,算法素养的内涵及要素有待进一步剖析.[研究设计/方法]基于"以人为核心AI"思想和当今聚焦个人发展的核心素养框架,立足多元主体视角界定算法素养的内涵及要素构成.[结论/发现]算法素养包括个体与算法交互的思维层面、态度层面和知识层面三个维度,且不同维度的构成要素因算法生产者、算法使用者和算法监管者等目标主体的改变而存在差异.未来可从理论研究、素养培育与算法治理三个方向深化算法素养的相关研究.[创新/价值]立足多元主体视角初步探索了算法素养的内涵,并针对算法生产者、算法使用者和算法监管者等多元主体提出了由思维、态度和知识层面组成的算法素养要素框架,为未来开展算法素养相关研究、加快全民算法素养提升、助力算法治理提供了清晰指引.

[目的/意义]开源软件社区用户通常从社区中获取代码相关知识,而缺乏贡献的动机和意愿,这将影响社区的可持续发展.[方法/过程]整合动机理论与社会资本理论,构建了开源社区用户知识贡献行为模型,采用混合方法包括SEM和fsQCA对数据进行分析.[结果/结论]研究发现,内部动机(流体验、自我效能)、外部动机(感知声誉、互惠)、社会互动关系、社区认同、共同语言显著影响用户知识贡献意愿和行为.fsQCA结果显示,流体验、感知声誉、互惠、信任是 4 个组态的共同核心条件.研究结果启示,开源软件社区需要关注用户的内外部动机,发展社会资本,以激发用户的知识贡献意愿和行为,促进开源社区持续快速发展.

[Purpose/Significance] Building an algorithm literacy evaluation index system is to provide a reference basis for strengthening the algorithm literacy education and improving the algorithm literacy level. [Method/Process] Using the literature research method, combined with the relevant research on algorithm literacy and literacy evaluation system, this paper constructed an algorithm literacy evaluation index system, carried out a questionnaire survey on the algorithm literacy of college students, and constructed an algorithm literacy evaluation model based on entropy weight TOPSIS method for empirical analysis. [Result/Conclusion] The empirical results show that there are significant differences in algorithm literacy between different grades and some disciplines. In addition, college students have stronger literacy ability and less inter-individual differences in algorithm awareness and algorithm social norms, while they have the opposite in algorithm knowledge and skills and critical thinking.

[Purpose/Significance] The risks derived from algorithmic recommendation services have a significant impact on people’s daily lives. Clarifying the generation mechanism of algorithmic recommendation service risks and the behavior characteristics of users’ response to algorithmic recommendation service risks will help to further improve the algorithmic recommendation service system. [Method/Process] Based on the grounded theory and semi-structured interview, this study conducted in-depth one-on-one interviews with 30 Internet users. Combined with the protective action decision model and coping behavior theory, this study explored the users’ coping mechanism, influence factors and decision paths of algorithmic recommendation service risks. [Result/Conclusion]The study finds that risk attributes, communication channels, platform characteristics and users’ behavior characteristics will have an impact on the spread of algorithm recommendation service risks. Based on the cognition of risk communication, users will form the perception of risk, stakeholder and protective behavior. Then, users will form attitudes and effect evaluation on the basis of perception, and make decisions and adopt different coping behaviors based on this. Ultimately, this study puts forward suggestions from the three levels of algorithm literacy, risk participation and individual learning to deal with the risk of algorithm recommendation service.

[目的/意义]新冠肺炎疫情期间虚假健康信息泛滥,参与虚假健康信息治理是高校图书馆保障公众信息需求、提供健康信息服务的重要手段.[方法/过程]运用网络调研与文献调研的方法,选取国外15所高校图书馆开展调研,从虚假健康信息辨别方法提供、多元信息素养教育开展、可靠健康信息资源建设、疫情相关研究支撑提供、疫情记忆档案库构建实施5个方面归纳国外高校图书馆的治理举措.[结果/结论]国内高校图书馆可以从重视和参与虚假健康信息治理、构建协同治理虚假健康信息体系、创新虚假健康信息治理服务3个方面参与虚假健康信息治理.

[目的/意义]研究旨在揭示社会化问答平台用户的养生健康信息需求分布特征,并深入探究需求产生的动机及演化趋势.[方法/过程]本文以社会化问答平台"知乎"中13万条养生问答数据作为研究对象,通过LDA模型提取需求话题,在离散时间序列基础上结合马斯洛需求层次理论对话题的关注度与关注热点进行演化分析.[结果/结论]用户养生信息需求涵盖20个话题;相比传统健康信息需求对疾病的聚焦,养生健康信息需求在内容上更多样,需求层次更高.需求的关注度演化上,安全需求与尊重需求成为热点,新冠疫情加强了用户对养生健康信息需求的关注.话题间的内在联系上,用户对尊重需求话题的关注度以"商品化"的形式转移至安全需求话题.[创新/局限]本文首次聚焦养生健康信息需求,通过话题与演化分析细粒度地挖掘用户养生健康信息需求的变化趋势.此外,本文数据源来自同一平台,后续研究可分析多平台用户的养生健康信息需求并对动机进行深化.

[目的/意义]鼓励在线健康社区医生的知识贡献行为是提升在线健康服务质量、推动在线健康社区可持续发展的重要保障.[方法/过程]文章以好大夫在线为研究对象,采用收益和成本视角,基于模糊集定性比较分析方法(fsQCA)对影响医生知识贡献行为的因素组态进行探究.[结果/结论]结果显示,触发在线健康社区医生知识贡献行为的影响因素组态可分为4类:荣誉收益型、物质收益型、物质与荣誉收益兼顾型以及低成本型,且荣誉收益是影响在线健康社区医生知识贡献行为的核心因素.揭示不同收益和成本因素组态的联合影响机制和实现路径,对在线健康社区提升医生的知识贡献活跃度和参与度具有一定的实践意义.