A Hybrid Model for Experts Finding in Community Question Answering

As a means to share knowledge, the community question answering (CQA) service provides users a chance to obtain or provide help by raising or answering questions. After a question is posted, the system must find an appropriate individual to answer this question. Several approaches have recently been proposed to find experts in CQA. In this paper, a new method to find experts in CQA is proposed by considering user post contents, answer votes, ratio of best answers, and user relation. The votes are used in post relation analysis to calculate user authority. The user´s knowledge score can be calculated through topic analysis. Considering that a question usually includes many trivial words, an accurate distribution is nearly impossible to obtain with LDA. To solve this problem, vocabulary is extended by including the link information shown in a question, the top 10 relevant words from Wikipedia are provided for each tag. Tag-LDA models the user topic distribution and predicts the topic distribution of new questions. An experiment is conducted on Stack Overflow dataset, which is the world´s largest computer programming CQA site. Experimental results showed approximately 2.97% to 7.79% performance improvement in nDCG@N metrics.