User intimacy model for question recommendation in community question answering

摘要

In this paper, we address the problem of automatic recommendation of new questions to suitable users in community question answering (CQA). The major challenge is the accurate selection of suitable users to answer a given question. Most approaches seek suitable users for a question by estimating their capability, interests or a blend of both. However, this ignores intimacy between the user and the asker of a question over different topics. Intimacy between askers and answerers is an important factor in question recommendation. For example, a user is likely to post an answer if interested in a question and intimate with its asker. We propose to model and learn intimacy between users over topics with social interaction in CQA for question recommendation using a novel topic model. We believe this paper is the first to estimate the intimacy between users over different topics and investigate influences on the performance of question recommendation in CQA. We propose a user intimacy model (UIM), an LDA-style model that incorporates social interaction in the generative process of a question-answer (QA) pair to model and learn intimacy between users over topics. Experiments using real-world data from Stack Overflow show that our UIM-based approach consistently and significantly improves the performance of question recommendation, demonstrating that our approach can increase question recommendation accuracy in CQA by utilizing the intimacy between users over topics and that this is an important factor in question recommendation.