Collaborative Filtering Using Explicit and Implicit Ratings for Arabic Dataset

Abstract

As the amount of digital information recorded on the internet increases, the need for flexible recommender systems is growing. Collaborative Filtering (CF) has been widely used in the E-commerce industry. A variety of input data was used, either implicitly or explicitly, to provide personalized recommendations for specific users and helped the system to improve its performance. Traditional CF algorithms relied solely on users' numeric ratings to identify user preferences. The majority of current research in recommender systems is focusing on a single implicit or explicit rating. In this paper, we combine explicit rating and implicit rating for user reviews to build the best recommender system using a large Arabic dataset. In addition, we employ two powerful techniques in the creation of our recommender system. First, we use Item-based CF and use cosine vector similarity to calculate the similarity between items. Second, we use Singular Value Decomposition (SVD) to reduce dimensionality, boost efficiency, and solve scalability and sparsity problems in CF. The proposed approach improves the experiment results by reducing mean absolute and root mean squared errors. The experimental results show to perform better when using both explicit and implicit ratings compared with using only one type of ratings. Keywords— Collaborative filtering (CF) Explicit and Implicit Ratings, A Large-Scale Arabic Book Reviews (LABR), LABR Lexicon.