Emotion-Based Movie Recommendations: How Far Can We Take This?

One of the underlying targets of movies is to evoke emotions in their viewers. When the viewers connect emotionally to a movie scenes and characters, then the watching experience naturally becomes stronger and more memorable. The importance of emotions is recognised by the movie industry and pre-screening sessions are regularly held to figure out if a movie reaches the desired emotional impact. Although there exists a large body of research on movie recommendations, the established recommendation techniques generally overlook the emotional aspects of movie watching. Collaborative methods rely on statistical correlations between movie ratings, whereas content-based and conversational methods primarily exploit implicit or explicit user preferences towards certain content features of movies. In this talk, we propose that recent developments in sensing technologies, affective computing, and data mining pave the way for a new generation of emotion-based movie recommender systems. We argue that the emotions evoked over the course of watching a movie offer a fertile ground for user modelling and movie recommendations. The evoked emotions can be considered as implicit user feedback regarding the emotional triggers planted in the movie by its creators. Clearly, users may respond in different ways to the same triggers, providing detailed information on their affective state and personality. This emotional information may be substantially richer than traditional ratings or feature preference scores typically exploited by recommender systems. Hence, it can fuel the delivery of better movie recommendations. We will discuss several challenges referring to the use of emotions in movie recommendations. How to accurately and unobtrusively capture user emotions evoked while watching a movie? What techniques can reliably interpret the captured emotions in context of the movie content and its emotional triggers? How can the information on the evoked user emotions and affective state augment the movie recommendation task? How to integrate these emotion-based recommendations with other recommendation techniques? Many of these questions are only partially answered and call for future research. Last but not the least, an important question to consider refers to the broad applicability of emotion-based personalisation. Emotional aspects of item consumption or purchase are not limited to movie watching, and similar challenges are faced in the music, retailing, advertisement, and other domains. Taking the emotion-based user modelling and recommendation beyond movie recommendations has a strong untapped potential both in the research and business worlds.