Expert recommendations with temporal dynamics of user interest in CQA

Community question answering (CQA) has become an essential service of promoting knowledge sharing in social platforms. To make question answering more efficient, several expert recommendation methods for CQA have been proposed, but most of them focus on the similarity matching between user interest and question content while ignoring the temporal dynamics of user interest, whose changes may decrease the quality of recommendation results. In this paper, a long and short term-based expert recommendation model (LSTERM) via attention mechanism-based CNN and Bi-GRU, which considers not only user interest but also user expertise, is proposed. The model can learn the embedded user/question feature representation from various content information by using attention mechanism-based CNN and then track the change of user interest and expertise over time by using Bi-GRU. Experiment results on real data demonstrate that with temporal dynamics, the recommendation accuracy is substantially improved compared with other state-of-the-art methods. Copyright © 2021 Springer Nature Switzerland AG.