Exploiting User Activities for Answer Ranking in Q&A Forums

Many Q&A forums suffer high variance in the quality of their contents because of their loose edit control. To solve this problem, many methods are proposed to rank answers based on their quality. Most existing works in this domain focus on using variable features or employing machine learning techniques to automatically assess the quality of answers. Few of these works noticed that the relationship formed by user’s activities can be helpful in capture the expertise of users in a specific topic. In this paper, we consider the relationship between users’s activities in answer ranking task, create three new topic-aware features based on user profile information and the network formed by user’s question-answering and comment activities, then we combine new created features with texture, user, comment features together and adopt a pairwise L2R approach SVMRank to rank answers. Experiments on a dataset extracted from Stack Overflow show that, (a) the new created features can better capture the expertise of users than other user features in answer ranking task. (b) the answer ranking approach get better performance when adding our new created features to the features used in previous works.