Conflict structure of movies as a predictor of preference similarity between movies

• Main Author: Ling, Audrey Yan-Hui
• Other Authors: Poong Oh
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Social sciences::Communication; Social sciences::Mass media::Media economics
• Online Access: https://hdl.handle.net/10356/147141
• Institution: Nanyang Technological University

In this study, we propose a novel method of predicting preference similarity for movies by focusing on their narrative structures. Based on Heider’s balance theory and the generalized concept of structural balance, we hypothesize that the similarity of conflict structure portrayed in movies will predict preference similarity between them. To test our hypothesis, we collected 1,122 movie scripts from the Internet Movie Script Database and extracted pairwise interactions among the characters in each movie using text analytics. We determined their relational valence, constructed a signed network for each movie, and finally, quantified the magnitude and complexity of conflicts among characters using computational methods. Preference similarity between movies was measured by analysing movie rating scores from 264,689 people. We examined how well the similarity of conflict structure predicts movie preference similarity using hierarchical regression analysis. Our results suggest that the network analysis of fictional characters provides a valid and reliable computational method to characterize movies and other unstructured textual data. This holds important implications for video streaming platforms who seek to develop their own recommender systems without the costs and requirements of more sophisticated tools such as collaborative filtering.