Deceptive Answer Prediction with User Preference Graph

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 1723–1732, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Deceptive Answer Prediction with User Preference Graph Fangtao Li§, Yang Gao†, Shuchang Zhou§‡, Xiance Si§, and Decheng Dai§ §Google Research, Mountain View ‡State Key Laboratory of Computer Architecture, Institute of Computing Technology, CAS {lifangtao,georgezhou,sxc,decheng}@google.com †Department of Computer Science and Technology, Tsinghua University gao young@163.com Abstract In Community question answering (QA) sites, malicious users may provide decep- tive answers to promote their products or services. It is important to identify and fil- ter out these deceptive answers. In this paper, we first solve this problem with the traditional supervised learning meth- ods. Two kinds of features, including tex- tual and contextual features, are investi- gated for this task. We further propose to exploit the user relationships to identify the deceptive answers, based on the hy- pothesis that similar users will have simi- lar behaviors to post deceptive or authentic answers. To measure the user similarity, we propose a new user preference graph based on the answer preference expressed by users, such as “helpful” voting and “best answer” selection. The user prefer- ence graph is incorporated into traditional supervised learning framework with the graph regularization technique. The ex- periment results demonstrate that the user preference graph can indeed help improve the performance of deceptive answer pre- diction. 1 Introduction Currently, Community QA sites, such as Yahoo! Answers1 and WikiAnswers2, have become one of the most important information acquisition meth- ods. In addition to the general-purpose web search engines, the Community QA sites have emerged as popular, and often effective, means of information seeking on the web. By posting questions for other participants to answer, users can obtain answers to their specific questions. The Community QA 1http://answers.yahoo.com 2http://wiki.answers.com sites are growing rapidly in popularity. Currently there are hundreds of millions of answers and mil- lions of questions accumulated on the Community QA sites. These resources of past questions and answers are proving to be a valuable knowledge base. From the Community QA sites, users can di- rectly get the answers to meet some specific infor- mation need, rather than browse the list of returned documents to find the answers. Hence, in recent years, knowledge mining in Community QA sites has become a popular topic in the field of artifi- cial intelligence (Adamic et al., 2008; Wei et al., 2011). However, some answers may be deceptive. In the Community QA sites, there are millions of users each day. As the answers can guide the user’s behavior, some malicious users are moti- vated to give deceptive answers to promote their products or services. For example, if someone asks for recommendations about restaurants in the Community QA site, the malicious user may post a deceptive answer to promote the target restaurant. Indeed, because of lucrative financial rewards, in several Community QA sites, some business own- ers provide incentives for users to post deceptive answers for product promotion. There are at least two …

Content truncated to 500 words