On the Use of Synopsis-based Features for Film Genre Classification
Giuseppe Portolese, Valéria Delisandra Feltrim · 2018
Technological advancements and the interest of companies that operate in digital environments have made the categorization of mediatic products increasingly popular. This is often a multi-label scenario, where an item may be labeled with many categories. Most of the literature approach film genre classification as a mono-label task, usually relying on audio-visual features. In this paper we explore the use of text-based features extracted from film synopses for multi-label film genre classification. We experimented with 19 feature extraction approaches combined with 4 multi-label classifiers. Our experimental results show f1-scores of up to 54.8%, which are significantly higher than other similar studies presented in the literature.