Structures or texts? A dynamic gating method for expert finding in CQA services

Abstract

Expert finding plays an important role in community question answering websites. Previously, most works focused on assessing the user expertise scores mainly from their past question-answering semantic features. In this work, we propose a gating mechanism to dynamically combine structural and textual representations based on past question-answering behaviors. We also use some user activities including temporal behaviors as the features, which determine the gate values. We evaluate the performance of our method on the well-known question answering sites Stackexchange and Quora. Experiments show that our approach can improve the performance on expert finding tasks.