Utilizando Tropes em modelos de recomendação híbridos

Abstract

Recommendation systems (SR) have been widely studied in recent decades. The growth of the Internet and the consolidation of Web 2.0 have contributed to the emergence of various services such as social networks, blogs, collaborative platforms, among others, resulting in increased volume of information. This scenario has fostered the development of new research on how to use such information to mitigate limitations of SRs and improve their quality. The recommendation of movies became one of the most discussed topics in the literature about SRs. The industry also contributed to its popularity with the growth of streaming services such as Amazon, Netflix, iTunes, and Google Play. In such scenario, different sources of information in Web have been exploited to extract features to describe movies. The most common approaches use features such as genre information, movie direction, cast, etc. Other approaches attempt to characterize the story itself by means of information about the content of movies, its story structure, elements of narrative and characters. Such content can be represented by Tropes. Tropes are the elements that make up a fictional story found in movies, books, comics and other contents. In this work, we present a systematic study of Tropes, investigating its relevance to the context of a story and how they can be incorporated in Movie Recommender Systems. The experiments performed in this research suggest that hybrid models based on the combination of tropes with the films genres can improve the precision of the predictions about 3% in comparison to traditional methods.