On dynamicity of expert finding in community question answering.

Community Question Answering is one of the valuable information resources which provide users with a platform to share their knowledge. Finding potential experts in CQA is beneficial to several problems like low participation rate of the users, long waiting time to receive answers and to the low quality of answers. Many research papers focused on retrieving the expert users of CQAs. Most of them are taking expertise into consideration at the query time and ignore the temporal aspects of the expert finding problem. However, considering the evolution of personal expertise over time can improve the quality of expert finding. In many applications, it is beneficial to find the potential experts in future. The proper identification of potential experts in CQA can improve their skills and the overall user participation and engagement. Considering dynamic aspects of the expert finding problem, we introduce the new problem of Future Expert Finding in this paper.Here, given the expertise evidence in current time, we aim to predict the best ranking of experts in future. We proposed a learning framework to predict such ranking on StackOverflow which is currently one of the most successful CQAs. We examine the impact of various features to predict the probability of becoming an expert user in future time. Specifically, we consider four feature groups; namely, topic similarity, emerging topics, user behavior and topic transition. The experimental results indicate the efficiency of the proposed models in comparison with several baseline models. Our experiments show that the performance of our proposed models can improve the MAP measure up to 39.7% in comparison with our best baseline method. Moreover, we found that among all of these feature groups, user behaviors have the most influence in the estimation of future expertise probability.