Personality in Recommender Systems

Abstract

The personality-based recommender systems (RS) has emerged as a new type of RS in recent years, given that personality contains valuable information enabling systems to better understand users' preferences [7]. This presentation first gives an overview of the state-of-the-art in this area, including the approaches developed for enhancing collaborative filtering (CF) by computing users' or items' personality similarity [1,4,5,8], as well as the one that incorporates personality into matrix factorization to predict items that users are able to rate for active learning [3].We then discuss several open issues. One issue is how to utilize personality to improve recommendation diversity. Diversity refers to the system's ability in returning different items in one set, which may help users more effectively explore the product space and discover unexpected items [6]. Our recent studies identified the effect of personality on users' diversity differences [2], and demonstrated that people perceive the system, which considers personality in adjusting recommendations' diversity degree, more competent and satisfying [9].We also show how to acquire personality through unobtrusive and implicit way, so as to save users' efforts in answering personality quizzes. Through testing an inference model in movie domain that unifies both types of domain-dependent and -independent features for deriving users' personality from their behavior, we proved that the implicitly inferred personality can also be helpful to augment the system's recommendation accuracy [10].Other open issues include how to develop personality-based cross domain RS for addressing the critical cold-start problem, how to exploit the influence of personality on users' emotions for boosting context-aware RS, and how to elicit more domain-independent features for generalizing the personality inference procedure.