Contextual-Boosted Deep Neural Collaborative Filtering Approach for Arabic Textual Documents Recommendation

Contextual-Boosted Deep Neural Collaborative Filtering Approach for Arabic Textual Documents Recommendation Ons Meddeb1,2 Mohsen Maraoui2 Mounir Zrigui2 1University of Sousse, Higher Institute of Computer Science and Communication Techniques ISITCom, Hammam Sousse 4011, Tunisia 2University of Monastir, Research Laboratory in Algebra, Numbers Theory and Intelligent Systems RLANTIS, Monastir 5000, Tunisia meddeb.ons@gmail.com, maraoui.mohsen@gmail.com, mounir.zrigui@fsm.rnu.tn Abstract The technological advancement and the expansion of using the internet has made it possible to communicate between people and machines. Many exchanged textual resources have been given simultaneously make users have difficulty in choosing the most appropriate items that will improve their knowledge. To deal with data sparsity problem, review-based recommendation systems have shown potential in a wide range of Natural Language Processing (NLP) tasks. Due to the high-dimensionality and the complex semantics of Arabic textual data, review-based approach is proposed for documents recommendation. Two parallel neural networks are introduced to learn items properties and users’ behaviors. Indeed, contextualized word representation model is used. Then, Gated Recurrent Units (GRU) is attached for extracting high-level-semantic features. After that, a shared layer modeled complex interactions between the latent vectors of users and items to improve subsequently rating prediction. Experiments showed the superiority of our proposed model compared with state-of-the-art methods. 1 Introduction Many technological advances have made it possible to communicate between peoples and machines (Slimi et al., 2022). The greatest benefit of delivering resources provided users with an opportunity to try a new learning style and explore its advantages (Hazar et al., 2022). With the increasing abundance of data over the internet, research on high-quality recommendation systems has gained interest in industry and academic. It has become very important for online platforms (e.g. e-commerce, music, books, social media, advertising, etc.) to provide the users what they need and like without wasting time searching (Dhelim et al., 2022). Most early recommendation systems use Collaborative Filtering methods (CF) that focus on learning accurate representations of user preferences and item features. Previous models used numeric ratings given by users as inputs. However, they suffered from data sparsity problem (Duan et al., 2022). They had difficulties for learning representations and generating reliable recommendations for users or items with few ratings. Another drawback of CF is that it did not make full use of the available context information like item attributes or user profile to make thereafter recommendations. To alleviate these problems, the use of textual reviews has attracted growing attention. Users can explain their opinions underlying their given ratings. They contain valuable and rich information that cannot be obtained from ratings alone (Jian et al., 2022). Most of reviews-based models have been focused on English language but not on Arabic. This is due to the richness of Arabic specificities (Meddeb et al., 2021a): a review can include different forms in the vocabulary or syntactic and semantic representations. For example, a word can have more than one lexical category in different contexts what changes the meaning of the sentence (Mahmoud and Zrigui, 2017). In this context, reviews-based recommendation models have attracted attention in …

Content truncated to 500 words