Deep Learning based Semantic Approach for Arabic Textual Documents Recommendation

Internet revolution has given readers access to vast and inexpensive resources with little effort. To deal with the proliferation of the exchanged data, textual documents recommendation has become a growing research topic of artificial intelligence and Natural Language Processing (NLP). It provided necessary guidance to choose the most suitable items. However, traditional rating based-methods had limits in terms of data sparsity. To address this lack, users’ reviews are used as a valuable source of knowledge. In this paper, deep learning based semantic approach is proposed for Arabic textual documents recommendation. This language received less attention compared to others because of its analysis complexity. Indeed, documents features from the reviews written for the item and user behaviors from the reviews written by the user are jointly modeled using deep models. The effectiveness of feedforward and recurrent neural architectures are studied to provide latent semantic representations. These features are further fed to a shared semantic space to capture complex user-item relations and predict finally ratings. Using an Arabic dataset, the experimental results demonstrated that the proposed approach was feasible to automate the process of recommending the most relevant Arabic textual documents.