基于信任理论与机器学习的在线评价质量评估模型研究

Our society is increasingly relying on the digitized opinions of others to make decisions in various online communities. However, advisors can freely express their opinions with little administration, and the quality of opinions may then vary. On the basis of trust theory, of all the possible reasons that result in diversity of opinion quality, the following two are the most dominant: 1) due to self-interest or other related motivations, some users might be dishonest and lie about their experience with entities; and 2) users are subjectively different, which thus leads to discrepancy of users’ opinions towards same entities. Correspondingly, online opinions may intentionally or unintentionally mislead the decision making of users. Therefore, the main objectives of this project are four-folds: 1) on the basis of trust theory, to identify the critical factors influencing the quality of online opinions from both the perspectives of motivations of advisors and subjectivity difference between users and advisors; 2) considering different information sources and different kinds of models, to construct an integrated trust model by leveraging artificial intelligence techniques (e.g., machine learning); 3) to construct a testbed for evaluating the effectiveness and robustness of the proposed trust model; and 4) to design controlled lab experiments to further validate the value of the trust model in real online communities. Our project addresses the limitations of previous work, including a lack of theoretical supports, poor explanatory power, and ineffective validation mechanism. Furthermore, the major expected output of this project, will not only provide theoretical support for research in both quality evaluation of online opinions and computational trust models, but also yield managerial insights for companies to construct better online opinions systems.

人们愈加依赖和重视来自在线社区中的他人评价制定日常生活中的各种决策。然而，因为缺乏有效的监管机制，在线评价质量问题愈来愈严重。由于评价者评价动机（利益驱动导致的不诚信等）以及主观性差异等因素，在线评价将有意或无意误导用户的决策过程。为此，本项目的研究目标为：1）基于信任理论，从评价动机和主观差异性两个方面确定影响在线评价质量的关键因素；2）采用以机器学习为代表的人工智能技术构建整合不同模型和不同信息格式的信任模型；3）设计相应的试验床环境检验信任模型对于在线评价质量评估的有效性和鲁棒性；4）设计控制实验进一步验证信任模型在实际系统中的应用价值。本项目的研究特色为：克服了以往信任模型缺乏理论支持、解释性和有效检验机制的缺点；其预期研究结果，不仅将为在线评价质量评估提供理论基础，进一步推进计算信任模型研究领域的发展，也将为实际在线社区的在线评价机制的建设提供有价值的管理启示。

本项目完成的主要研究内容有：1）基于信任理论，确定质量关键影响因素；2）整合不同数据源的模型；3）构建模型评价的实验床，促进模型比较的开放性和公平性；4）验证系统的实际价值。在本项目的资助下，我们基本完成了四个方面的内容，发表8篇研究成果（其中包括权威SCI/SSCI 2篇，人工智能顶级会议AAAI 2篇，信息系统顶级会议ICIS 1篇）。项目组培养了相关的硕士研究生5人，均已毕业。博士生1人，即将毕业。课题组参加了6次国际会议来宣传研究成果、以及1次国际访问来推进项目。