A Novel Diversity Measure for Understanding Movie Ranks in Movie Collaboration Networks

Abstract

We are interested in the relationship between the team composition and the outcome in the filmmaking process. We studied the “diversity” of the group of actors and directors and how it is related to the movie rank given by the audience. The “diversity” is considered as the representation of the degree of variety based on the possibilities of collaborations among its actors and directors. Their collaboration network for the movie was first generated from the “background” network of the collaborations from other works. Then a shortest-path method together with the Adamic/Adar method are used to form indirect links. Finally the “complete” collaboration network can be generated and the “diversity” measures are thus defined accordingly. We experimented on the France and Germany datasets and identified consistent patterns: the lower the “diversity” is, the lower the movie rank will be. We also demonstrated that a subset of our diversity measures were effective in the binary classification task for movie ranks, while the advantages are prone to Precision/Recall depending on the specific dataset. This further shows that the “diversity” measure is feasible and effective in distinguishing movie ranks.