Topic sensitive hybrid expertise retrieval system in community question answering services

Abstract

Here, we propose a topic sensitive hybrid expertise retrieval system in community question answering services. We introduce three new expertise signatures: knowledge, reputation, and authority. These signatures consider the questions, and hence, their answerers from a topic sensitive perspective. We estimate the knowledge of an answerer on a new question based on the previously answered subset of questions with similar topic distributions to the new question. The reputation of an answerer, moreover, is derived from the qualities of previously answered questions by the answerer with similar distributions of topics. Furthermore, we propose a topic sensitive authority model. It considers some topic related information associated with questions and the relationships among their answerers. We compare the proposed method with 26 existing methods on 4 real-world datasets using 5 performance measures. It outperforms the comparing algorithms in 91.73% (477 out of 520) cases.