Novelty In Recommender Systems for Effective Personalization in E-Commerce and Retail

Abstract

When trying to make recommendations to the users by leveraging users' past behavior, Recommender Systems generate item suggestions tailored to their preferences. However, users may become disenchanted with repetitive recommendations over time. To address this challenge, the inclusion of novel items becomes crucial. Novelty pertains to introducing users to previously unknown items, while serendipity adds an element of surprise to recommendations, ultimately enhancing user satisfaction. It is important to note that serendipity, due to its subjective nature, lacks a universally accepted definition. Consequently, this paper explores various perspectives on defining serendipity and examines techniques like deep learning to facilitate its incorporation into recommendation algorithms. The Social Choice Theory encompasses the study of aggregating individual preferences to make collective decisions. When incorporating Collaborative Filtering into recommendation systems, the principles of Social Choice Theory are employed. To evaluate the efficacy of a specific algorithm in delivering serendipitous recommendations, this study utilizes the Serendipity 2018 dataset, provided by the MovieLens research group. By comparing the algorithm's performance against that of other baseline algorithms, an assessment is made regarding its ability to introduce surprise elements and enhance user satisfaction. Analyzing the results offers valuable insights into the algorithm's effectiveness in generating serendipitous recommendations, thereby contributing to the advancement of recommendation systems.