Community-aware ranking algorithms for expert identification in question-answer forums

Abstract

Question-Answer forums (QAF) are significant platforms for disseminating informal information and play important role in problem solving and learning. Expert identification still has some limitations and link analysis methods do not consider community dimension. In this paper an authority analysis approach for identifying experts is proposed. This approach combines overlapping community detection (OCD) algorithms with ranking methods to compute the nodes' expertise level in QAFs. Firstly, graph resulting from a specific search query is computed and an OCD algorithm is applied on it. After identifying clusters of nodes, we change updating rules of original Hyperlink-Induced Topic Search (HITS) and PageRank to take the effect of intra cluster links and extra cluster connections. People whom are intra or overlapping to a community possess higher vision about context of the community than nodes which are outside. We experimented the proposed overlapping community-aware ranking algorithms and compared them with baseline approaches on online forums. Results indicate that OCD improves expert identification accuracy and relevancy.