Curiosity-inspired Personalized Recommendation

Pranita Shrestha, Min Zhang, Yiqun Liu, Shaoping Ma

Abstract: Historically, research on better recommendations revolved mainly around feeding the user with accurate recommendations. However, now the spectrum has shifted from the conventional idea of focusing only on accuracy, to focusing on other dimensions like serendipity, novelty, and diversity, and creating an optimal balance to increase the users' satisfaction and experience. This paper explores the realm of novelty and curiosity through two of the most prominent psychological concepts of curiosity - Wundt Curve and Berlyne Psychology. Through this exploration, we propose a computational framework called Curiosity-inspired Personalized Recommendation re-ranking framework. This model considers both accuracy as well as personalized levels of novelty and curiosity while generating the final recommendation list. It is flexible to work with multiple traditional algorithms to create a joint optimization approach to provide the user with the desired level of novelty and curiosity, from which they can thrive. Comparative experiments have been made to evaluate the proposed model in terms of various evaluation metrics (accuracy, ranking, novelty, and curiosity metrics). Results show that it is possible to recommend novel items by causing very little to no accuracy loss. Sometimes, the model even manages to improve accuracy. This paper shows that psychological theory related to curiosity helps personalized recommendations in balancing and improving both novelty and accuracy. This, in turn, sheds light upon interdisciplinary research in the news streaming domain.