Multilingual Sentiment Recommendation System based on Multilayer Convolutional Neural Networks (MCNN) and Collaborative Filtering based Multistage Deep Neural Network Models (CFMDNN)

Abstract

Recommendation systems referred to synonymously as Recommender systems as well, are one of the most important computer-based online computational platforms, used in retail and business sector extensively. They play an important role in providing filtered and selective information that is useful to customers and can help in both enhancing the user experience with the platform as well as increase the prospect of purchase of goods or provision of services. Traditional approaches for recommendation systems are based on either collaborative filtering technique, or content-based filtering technique or a hybrid combination of both these techniques. However, these are certain limitations to these traditional approaches, including the cold start problems, and the need for prior user history, and interaction with items or service provided by the platform. In this paper we propose a novel computation framework based on use of deep learning models, along with inclusion of both explicit feedback in terms of ratings score and sentiments expressed in textual comments, for addressing the limitations of traditional recommender system approaches. The proposed deep learning framework allows joint modelling between users, items, and multiple channels of explicit feedback, with two novel deep neural models, the Collaborative Filtering based Deep Neural Network architecture (CFMDNN) model and the novel Multichannel Convolutional Neural Network (MCNN) model, for aiding the traditional Collaborative Filtering based system, and improve the performance of recommendation systems' performance. The evaluation of the proposed deep learning-based framework, in a multilingual context, based on two publicly available English and Arabic languages datasets, has shown promising outcomes in terms of ratings prediction accuracy used for assessing the performance.