Abstract
In this paper, we approach the Arabic community question answering problem. We integrated different types of similarity features, in addition to exploring the effect of using preprocessing. Moreover, we developed a novel deep neural network ensemble model that outperformed the previously achieved performance. Our ensemble model benefits from the semantic and lexical similarity features. Also, our ensemble model uses the recent advances in language models using BERT model. Our proposed model produced a MAP value of (62.9%), an AvgRec value of (86.6%), and an MRR value of (68.86%). (C) 2020 The Authors. Published by Elsevier B.V. on behalf of Faculty of Engineering, Alexandria University.