Learning Character Network Features for Movie Genre Classification

Abstract

This paper investigates the effectiveness of character network features for movie genre classification. We construct character networks for each movie in the Movie Galaxy dataset, capturing the interaction patterns between characters. We extract three types of features from these networks: topological features describing the network structure, spectral features derived from the network's eigenvalues, and embedding features learned using dimensionality reduction techniques. These features are then fed into four machine learning classifiers: Random Forest, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Gaussian Process Classifiers (GPC). We evaluate the performance of each classifier on the movie genre classification task. Our results demonstrate that SVM and GPC achieve the best accuracy, suggesting that character network features hold promise for movie genre prediction.