Incorporate Credibility into Context for the Best Social Media Answers

Incorporate Credibility into Context for the Best Social Media Answers Qi Su a,b, Helen Kai-yun Chen a, and Chu-Ren Huang a aDepartment of Chinese & Bilingual Studies, The Hong Kong Polytechnic University, Hong Kong, China sukia@pku.edu.cn bKey Laboratory of Computational Linguistics, Peking University, Beijing, China {helenkychen, churenhuang}@gmail.com Abstract. In this paper, we focus on the task of identifying the best answer for a user- generated question in Collaborative Question Answering (CQA) services. Given that most existing research on CQA has focused on non-textual features such as click-through counts which are relatively difficult to access, we examine the effectiveness of diverse content-based features for the task. Specially, we propose to explore how the information of evidentiality can contribute to the task. By the comparison of diverse textual features and their combinations, the current study provides useful insight into the issues of detecting the best answer to a given question in CQA without user features or system specific link structures. Keywords: collaborative question answering, answer assessment, credibility, quality 1 Introduction The technology of web search has provided a powerful platform for seeking and accessing information on the Web. However, due to the limitations of keyword based retrieval, people often find it hard to locate their desired content through ad hoc searching. Therefore, it is no surprise that Web 2.0 system is trying to seek improvement in information foraging. Given that a general- purpose, fully-automated question answering is still beyond the state-of-the-art, the human- powered Collaborative Question Answering (CQA) services start to draw new attentions. Through CQA services, people could post their questions concerning any subject, even those the ad hoc searching may find difficult to answer, on a community portal, and get answers directly provided by community contributors. Due to the advantage of human interaction, CQA has become a popular question-answering platform and an active research area up till now. It turns out to be an effective supplement to the ad hoc searching and question answering (Banerjee and Han, 2009; Weerkamp and Rijke, 2008). Although CQA tries to help users come up with the best answers with its human-oriented strategy, the quality of user-generated answers is sometimes difficult to control (Gyongyi et al, 2008). Among the answers received, there might be only few knowledgeable responses and helpful information. Some postings could be vague or even purposely misleading. Thus seeking the best answers to questions asked within the CQA community has been a challenge to the CQA system. The existing researches on best answer seeking in CQA can roughly be categorized by textual feature based and non-textual feature based approaches. Usually, best answers could be decoded as high-quality and authoritative. To date, there is no much attention on the content-based answer quality judgment in CQA research. Instead, non-textual features are utilized extensively in estimating the quality of answers (Jeon et al., 2006). Much of the related work is based on user authority analysis by the user’s interaction network. In some related area, researchers have also PACLIC 24 Proceedings 535 proposed textual feature based …

Content truncated to 500 words