Graph Prediction in Movie Graphs

Graphs can be used in applied in both an unsupervised and supervised context that can augment classification tasks. This paper focuses on a set of movie graphs that correspond to actor interactions in movies associated with various genres. I first compare the graph statistics between two disparate genres and perform random walks and analyze the eigenvalue and eigenvectors of the adjacency matrices. I then build predictive models based on the graph statistics and classify movies into genres with varying degrees of accuracies. Additionally, I obtain fingerprints associated with graphs by running the adjacency matrix through convolutional neural networks (CNNs) in order to classify movie graphs into genres with greater accuracy than relying only on graph statistics.