Deep Neural Network-Based Models for Ranking Question - Answering Pairs in Community Question Answering Systems

摘要

Ranking question-answering pairs according to their similarities to each input question is very important for any real-world community Question Answering system. To address this problem we will propose the models which use Convolutional Neural Network and Bi-Directional Long Short Term Memory. The proposed models are formulated for both representation learning and question similarity score detection. Especially in this paper we will utilize various feature kinds including both abstract features (i.e. high level representation) and conventional features. We test our proposed model on the dataset SemEval 2016 and obtain the results with the Accuracy and MAP of 82.86% and 78.43% respectively, which are best in comparison with previous studies.