Detecting Spammers in Community Question Answering

International Joint Conference on Natural Language Processing, pages 118–126, Nagoya, Japan, 14-18 October 2013. Detecting Spammers in Community Question Answering Zhuoye Ding, Yeyun Gong, Yaqian Zhou, Qi Zhang, Xuanjing Huang Fudan University School of Computer Science {09110240024,12110240006,zhouyaqian,qz,xjhuang}@fudan.edu.cn Abstract As the popularity of Community Ques- tion Answering(CQA) increases, spam- ming activities also picked up in number- s and variety. On CQA sites, spammers often pretend to ask questions, and selec- t answers which were published by their partners or themselves as the best answers. These fake best answers cannot be easily detected by neither existing methods nor common users. In this paper, we address the issue of detecting spammers on CQA sites. We formulate the task as an opti- mization problem. Social information is incorporated by adding graph regulariza- tion constraints to the text-based predic- tor. To evaluate the proposed approach, we crawled a data set from a CQA portal. Experimental results demonstrate that the proposed method can achieve better per- formance than some state-of-the-art meth- ods. 1 Introduction Due to the massive growth of Web 2.0 technolo- gies, user-generated content has become a prima- ry source of various types of content. Communi- ty Question Answering (CQA) services have also attracted continuously growing interest. They al- low users to submit questions and answer ques- tions asked by other users. A huge number of users contributed enormous questions and answers on popular CQA sites such as Yahoo! Answers1, Baidu Zhidao2, Facebook Questions3, and so on. According to a statistic from Yahoo, Yahoo! An- swers receives more than 0.82 million questions 1http://answers.yahoo.com 2http://zhidao.baidu.com 3http://www.facebook.com and answers per day4. On CQA sites, users are primary contributors of content. The volunteer-driven mechanism brings many positive effects, including the rapid growth in size, great user experience, immediate response, and so on. However, the open access and reliance on users have also made these systems becoming targets of spammers. They post advertisements or other irrelevant answers aiming at spreading advertise or achieving other goals. Some spam- mers directly publish content to answer question- s asked by common users. Additionally, another kind of spammers (we refer them as “best answer spammers”) create multiple user accounts, and use some accounts to ask a question, the others to pro- vide answers which are selected as the best an- swers by themselves. They deliberately organize themselves in order to deceive readers. This kind of spammers are even more hazardous, since they are neither easily ignored nor identifiable by a hu- man reader. Google Confucius CQA system also reported that best answer spammers may generate amounts of fake best answers, which could have a non-trivial impact on the quality of machine learn- ing model (Si et al., 2010). With the increasing requirements, spammer de- tection has received considerable attentions, in- cluding e-mails(L.Gomes et al., 2007; C.Wu et al., 2005), web spammer (Cheng et al., 2011), review spammer (Lim et al., 2010; N.Jindal and B.Liu, 2008; ott et al., 2011), social media spammer (Zhu et al., 2012; Bosma et al., 2012; Wang, 2010). …

Content truncated to 500 words