Issues in Decision Tree Classification of Film Genre Using Plot Features

Classification and abstraction of narrative content is a problem that spans several domains and disciplines ranging from Film Studies and Narratology to Adaptive Hypermedia and Computational Preference Modeling. Applications for automatic recognition and classification of narrative content range from commercial recommendation systems to interactive digital storytelling (IDS) systems. The notion of genre is commonly applied to a narrative as a means of classification; however genre is imprecise for taxonomic classifications, as it suffers from various overlaps and inconsistencies that defy systematic categorization. This paper describes a machine learning approach to genre classification that uses a decision tree to learn the genre associations of films based on a database of “plot keywords”, representing narrative and stylistic features. While the approach is ultimately unsuccessful, the challenges encountered point at some important lessons for future work in computational genre recognition.