-
摘要: 信任计算作为一个热门课题已经在学术界广为研究。以往很多信任模型主要是基于声望机制的，例如Kamvar等人提出的EigenTrust模型，Yu等人提出的基于声望的信任管理机制，以及南加州大学的Kai Hwang等人提出的基于模糊逻辑推理的P2P声望系统等。这些研究工作往往都忽略了应用场景中大量的内容（content）和上下文（context）信息，因此会带来以下几个问题：1. 信任关系常常被泛化了；2. 显式信任关系矩阵往往比较稀疏；3. 难以刻画用户之间的信任程度；4. 算法的时间复杂度高。为了解决以上这些问题，本章提出了一个集成声望、内容和上下文信息的组合信任模型——RCCtrust（Reputation-, Content-, Context-based trust model），来支持更加准确、高效和细粒度的信任计算。
首先，我们从互联网的内容和上下文中抽取出信任相关的信息来扩展基于声望的信任机制（reputation-based trust mechanism）。然后，我们利用基于角色（role-based）和基于行为（behaviour-based）的信任推理功能推理出用户的兴趣爱好和领域相关的信任关系（category-specific trust relationship）。根据社会学领域的研究成果，RCCtrust利用相似度来刻画用户彼此之间的信任程度。最后我们构建出具体领域内的组合信任网络来进行信任计算。
实验中我们以商品推荐系统作为应用场景，利用Epinions网站上的真实在线社会网络数据进行实验。实验数据集是通过爬取Epinions网站上每个用户的个人主页获得的，我们从Media这个主类别中随机选取一个Top Reviewer，然后顺着他的信任关系和被信任关系的链接去寻找其他用户。我们抓取这些用户的角色、所撰写的review以及所提供的反馈评分，并根据前面介绍的方法推理出他们的兴趣爱好和领域相关的信任关系。同时我们从Epinions网站上抓取商品评分和反馈评分来计算成对用户之间的相似度。为了简单起见，在本实验中我们没有分析用户自我介绍等文本内容。
我们将RCCtrust同以往的两种方法进行比较：一种是传统协同过滤的单纯相似度方法PureSim，另一种是Massa和Avesani提出的仅利用信任关系的trust-aware方法。为了评估三种方法的性能，我们进行两组实验。第一组是关于商品评分预测的准确率。我们采用交叉验证方法（leave one out），并利用平均绝对误差（MAE）和用户平均绝对误差（MAUE）来度量预测错误率。第二组实验是关于评分预测的覆盖率。一个推荐系统不可能为每个用户和每种商品都能进行评分预测。因此，预测覆盖率也是衡量这些方法的一个重要指标。我们计算两种覆盖率：一是评分覆盖率，另一种是用户覆盖率。评分覆盖率指的是能够被预测出来的商品评分的百分比。而用户覆盖率是这样定义的：如果能为某个用户至少预测出一个商品评分，那么这个用户就是可覆盖的，可覆盖用户占全体用户的百分比就是用户覆盖率。我们在两个数据集上进行实验：第一个数据集是评分较多用户数据集H，H中包含了整个数据集中所有评分较多的用户。第二个数据集是评分较少用户数据集S，S中包含了整个数据集中所有只评价了零星几个商品的用户。
实验结果显示，RCCtrust模型无论在准确率还是覆盖率上都优于传统协同过滤的单纯相似度方法PureSim和仅仅利用信任关系的trust-aware方法。同时我们也看到RCCtrust模型具有较强的健壮性，不但在为评分活跃用户进行预测时能取得较好的结果，而且对于冷启动用户的评分预测也十分有效。
本文的主要贡献是提出了一个集成声望、内容和上下文信息的组合信任模型RCCtrust，给出了基于语义的信任推理机制推理出用户的兴趣爱好，从泛化的信任关系推理出领域相关的信任关系，以及挖掘出隐式信任关系。不同于以往的一些研究工作提出了一些信任框架和信任政策，RCCtrust是一个具体的信任计算模型，我们给出了具体的信任计算方法和相应的信任推理机制，并在真实的在线社会网络数据上进行了大量的实验，详尽的实验结果验证了RCCtrust模型的有效性。Abstract: Previous trust models are mainly focused on reputational mechanism based on explicit trust ratings. However, the large amount of user-generated content and community context published on Web is often ignored. Without enough information, there are several problems with previous trust models: first, they cannot determine in which field one user trusts in another, so many models assume that trust exists in all fields. Second some models are not able to delineate the variation of trust scales, therefore they regard each user trusts all his friends to the same extent. Third, since these models only focus on explicit trust ratings, so the trust matrix is very sparse. To solve these problems, we present RCCtrust --- a trust model which combines Reputation-, Content- and Context-based mechanisms to provide more accurate, fine-grained and efficient trust management for the electronic community. We extract trust-related information from user-generated content and community context from Web to extend reputation-based trust models. We introduce role-based and behavior-based reasoning functionalities to infer users' interests and \emph{category-specific} trust relationships. Following the study in sociology, RCCtrust exploits similarities between pairs of users to depict differentiated trust scales. The experimental results show that RCCtrust outperforms pure user similarity method and linear decay trust-aware technique in both accuracy and coverage for a Recommender System.-
Keywords:
- content-based /
- context-based /
- reputation-based /
- trust model /
- web of trust
-
-
[1] Zhang Y, Chen H, Wu Z, Zheng X. A reputation-chain trust model for the semantic Web. In Proc. the 20th International Conference on Advanced Information Networking and Applications, Vienna, Austria, April 18-20, 2006, pp.719-723.
[2] Golbeck J, Parsia B, Hendler J. Trust networks on the semantic Web. In Proc. Cooperative Intelligent Agents VII, Helsinki, Finland, August 27-29, 2003, pp.238-249.
[3] Hill S, Provost F, Volinsky C. Network-based marketing: Identifying likely adopters via consumer networks. Statistical Science, 2006, 21(2): 256-276.
[4] Epinions. http://www.epinions.com.
[5] Domingos P, Richardson M. Mining the network value of customers. In Proc. the Seventh ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, California, USA, August 26-29, 2001, pp.57-66.
[6] Richardson M, Domingos P. Mining knowledge-sharing sites for viral marketing. In Proc. the Eighth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Alberta, Canada, July 23-26, 2002, pp.61-70.
[7] Massa P, Avesani P. Trust-aware collaborativefiltering for recommender systems. In Proc. Federated International Conference On The Move to Meaningful Internet: CoopIS, DOA, ODBASE, Larnaca, Cyprus, October 25-29, 2004, pp.492- 508.
[8] Victor P, Cornelis C, Teredesai A M, De Cock M. Whom should I trust? The impact of keyfigures on cold start recommendations. In Proc. the 23rd Annual ACM Symposium on Applied Computing, Ceara, Brazil, March 16-20, 2008, pp.2014-2018.
[9] Kamvar S D, Schlosser M T, Garcia-Molina H. The Eigentrust algorithm for reputation management in P2P networks. In Proc. the Twelfth International World Wide Web Conference, Budapest, Hungary, May 20-24, 2003, pp.640-651.
[10] Wang Y, Vassileva J. Bayesian network-based trust model. In Proc. the 2003 IEEE/WIC International Conference on Web Intelligence, Halifax, Canada, October 13-16, 2003, pp.372- 378.
[11] Yu B, Singh M P. A social mechanism of reputation management in electronic communities. In Proc. the Fourth International Workshop on Cooperative Information Agents, Boston, USA, July 4-7, 2000, pp.154-165.
[12] Liang Z, Shi W. Pet: A personalized trust model with reputation and risk evaluation for P2P resource sharing. In Proc. the 38th Annual Hawaii International Conference of System Sciences, Big Island, USA, January 3-6, 2005, p.201b.
[13] Liang Z, Shi W. Enforcing cooperative resource sharing in untrusted P2P computing environments. Mobile Networks and Applications, 2005, 10(6): 971-983.
[14] Liang Z, Shi W. Analysis of ratings on trust inference in open environments. Performance Evaluation, 2008, 65(2): 99-128.
[15] Xiong L, Liu L. Peertrust: Supporting reputation-based trust for peer-to-peer electronic communities. IEEE Transactions on Knowledge and Data Engineering, 2004, 16(7): 843-857.
[16] Xiong L, Liu L. A reputation-based trust model for peer-topeer ecommerce communities. In Proc. IEEE International Conference on E-Commerce, Newport Beach, USA, June 24- 27, 2003, pp.275-284.
[17] Bizer C, Oldakowski R. Using contextand content-based trust policies on the semantic Web. In Proc. the 13th International Conference on World Wide Web, New York, USA, May 17-20, 2004, pp.228-229.
[18] Hu M, Liu B. Mining opinion features in customer reviews. In Proc. the Nineteenth National Conference on Artificial Intelligence, San Jose, USA, July 25-29, 2004, pp.755-760.
[19] Edelman trust barometer. November 2007. http://www.edelman.co.uk/trustbarometer/.
[20] Bellman R, Giertz M. On the analytic formalism of the theory of fuzzy sets. Information Sciences, 1973, 5: 149-156.
[21] Jean-Sebastien Brunner, Li Ma, Chen Wang et al. Explorations in the use of Semantic Web technologies for product information management. In Proc. the 16th International World Wide Web Conference, Banff, Canada, May 8- 12, 2007, pp.747-756.
[22] Rdf primer. February 2004, http://www.w3.org/TR/rdfprimer/.
[23] Chun Zeng, Chun-Xiao Xing, Li-Zhu Zhou, Xiao-Hui Zheng. Similarity measure and instance selection for collaborativefiltering. International Journal of Electronic Commerce, 2003, 8(4): 115-129.
计量
- 文章访问数: 26
- HTML全文浏览量: 0
- PDF下载量: 1944
下载:
