A Hybrid Neural Model for Identifying Similar Questions in Community Question Answering

Abstract Community Question Answering (cQA) platforms have emerged as valuable resources for knowledge sharing, where users submit questions and obtain answers from the community. A central task in cQA is to automatically identify semantically similar questions and rank them according to their relevance to a new input query. In this paper, we propose a hybrid neural model that integrates Bidirectional Long Short-Term Memory (BLSTM) networks with a Multi-Layer Perceptron (MLP) to jointly learn question representations and compute similarity scores. To further improve similarity estimation, the model also incorporates auxiliary external knowledge features, such as question type and category, in a lightweight manner. We evaluate our approach on two benchmark datasets, SemEval 2016 Task 3 and Quora Question pairs. Experimental results show that our hybrid model consistently outperforms strong baselines and achieves competitive performance compared to recent state-of-the-art methods. Keywords—community question answering, neural network, question similarity, external knowledge, BLSTM.