Bayesian Experts Recommendation in CQA: levereging aptitude, tag-based profiling, and social graph modeling

Expert recommendation plays a vital role in question-answering communities by ensuring that questions receive prompt and relevant responses. In this paper, we introduce PARE (aPtitude-bAsed Recommendation of Experts), a novel model-based approach designed to recommend expert repliers. The model uniquely integrates replying aptitude with social interactions among users, focusing on question tags to streamline the recommendation process without the need for extensive text analysis. Our method routes questions to the most suitable answerers by analyzing their past interactions with specific tags, while also predicting their aptitude for topics they have yet to cover. This is achieved through latent factor embeddings of users and tags, inferred under a fully Bayesian model that incorporates both tag-based replying patterns and the social graph of user interactions. Extensive experiments on real-world datasets show that PARE consistently outperforms baseline models across multiple evaluation metrics, proving its effectiveness in expert recommendation tasks.