Predicting Post Importance in Question Answer Forums Based on Topic-Wise User Expertise

Abstract

Q & A forums on the web are aplenty and the content produced through such crowd-sourced efforts is generally of good quality and highly beneficial to novices and experts alike. As the community matures, however, the explosion in the number of posts/answers leads to the information overload problem. Many a times users having expertise in a particular area are not able to address quality issues raised in the area maybe due to the positioning of the question in the list displayed to the user. A good mechanism to assess the quality of questions and to display it to the users depending on their area of expertise, if devised, may lead to a higher quality answers and faster resolutions to the questions posted. In this paper we present the results of our investigations into the effectiveness of various mechanisms to represent user expertise to estimate a post score reflecting its quality/utility of the post. We follow three different approaches to building a user profile representing the user’s areas of expertise: topic models based approach, tag-based approach and semantic user profiling approaches. We present the results of experiments performed on the popular Q&A Forum Stack Overflow, exploring the value add offered by these approaches. The preliminary experiments support our hypothesis that considering additional features in terms of user expertise does offer an increase in the classification accuracy even while ignoring features computable only after the first 24 hours. However, the proposed method to individually leverage on the semantic tag relations to construct an enhanced user profile did not prove beneficial.