Toward effective movie recommendations based on mise-en-scene film styles

Published in Proceedings of the 11th ACM Biannual Conference on Italian SIGCHI Chapter, 2015

Recommended citation: Yashar Deldjoo, Mehdi Elahi, Massimo Quadrana, Paolo Cremonesi Proceedings of the 11th ACM Biannual Conference on Italian SIGCHI Chapter (CHItaly 2015).

[DOI] [PDF] [bibtex]

Abstract

Recommender Systems (RSs) play an increasingly important role in video-on-demand web applications such as YouTube and Netflix characterized by a very large catalogs of videos and movies. Their goal is to filter information and to recommend to users only the videos that are likely of interest to them. Recommendations are traditionally generated on the basis of user’s preferences on movies’ attributes, such as genre, director, actors. Preferences on attributes are implicitly detected by analyzing the user’s past opinions on movies. However, we believe that the opinion of users on movies is better described in terms of the mise-en-scene, i.e., the design aspects of a film production affecting aesthetic and style. Lighting, colors, background, and movements are all examples of mise-en-scene. Although viewers may not consciously notice film style, it still affects the viewer’s experience of the film. The mise-en-scene highlights similarities in the narratives, as filmmakers typically relate the overall film style to reflect the story, and can be used to categorize movies at a finer level than with traditional film attributes, such as genre and cast. Two films may be from the same genre, but they can look different based on the film style. In this paper, we present an ongoing work for the development of a novel application that offers a personalized way to search for interesting multimedia content. Instead of using the traditional classifications of movies based on attributes such as genre and cast, we use aesthetic movie features derived from film styles as determined by filmmaker professionals. Stylistic features of movies are extracted with an automatic video analysis tool and are used by our application to generate personalized recommendations and to help users in searching for interesting content.