Enhancing Cross Domain Recommendation System Using Optimal Knowledge Transfer and Deep Learning Approaches

Abstract

The rapid growth of digital content platforms has made personalized newlinerecommendation systems crucial to enhancing user experience by suggesting relevant newlinemovies, books, and many other types of items. However, the traditional Recommender newlineSystem (RS) continues to face significant challenges in domains like movies, books and newlineliterature due to the sparsity of data users tend to give only a few ratings or interactions, newlinemaking it difficult to model their preferences precisely. Moreover, the transfer of newlineknowledge across various domains-for example, from movies to books-always turns out newlineto be inadequate because of data mismatch and inability to identify necessary newlinerelationships among items of user sequences; for example, a user interested in historical newlinenovels might watch a movie set in the same historical context. Furthermore, since the newlineinterests and preferences of users are continuously evolving, traditional newlinerecommendation systems cannot adapt to this dynamic behavior, yielding newlinerecommendations that are increasingly incorrect over time. Cross-Domain newlineRecommendation Systems (CDRS) is one of the feasible methods to overcome the newlineabove challenges by moving information from one domain to another domain to newlineimprove the accuracy of the recommendation system. However, several emerging newlinetechniques in cross-domain sequential recommendations often fail to effectively address newlinethe interconnected challenges of data scarcity, suboptimal knowledge transfer, and newlineshifting user preferences. In this thesis, CDRS is explored through optimal knowledge newlinetransfer and deep learning techniques and proposed three different methodologies to newlineaddress the problems in recommender systems like data sparsity, negative transferring newlinedue to domain discrepancies and dynamic drift in user preferences newline