Towards comprehensive expert finding with a hierarchical matching network

Abstract

In Community Question Answering (CQA) websites, expert finding aims to seek relevant experts for answering questions. The core of expert finding is to match candidate experts and target questions precisely. Most existing methods usually learn a single feature vector for the expert from the historically answered questions, and then match the target question, which would lose fine-grained and low-level semantic matching information. In this paper, instead of matching with a unified expert embedding, we propose an expert finding method with a multi-grained hierarchical matching framework, named EFHM. Specifically, we design a word-level and question-level match encoder to learn the fine-grained semantic matching between each historical answered question and target question, and then propose an expert-level match encoder to learn an overall expert feature for matching the target question. Through the hierarchical matching mechanism, our model has the potential to capture the comprehensive relevance between candidate experts and target questions. Experimental results on six real-world CQA datasets demonstrate that the proposed method could achieve better performance than existing state-of-the-art methods.