Comparison of Graph Distance Measures for Movie Similarity Using a Multilayer Network Model

Area of Science:

• Graph theory
• Network analysis
• Computational linguistics

Background:

• Graph distance measures are key for comparing graph structures.
• Movie network analysis traditionally uses single-layer models.
• A comprehensive multilayer network model is needed for richer story representation.

Purpose of the Study:

• To develop a multilayer network model for movie stories.
• To quantify movie story similarity using layer-to-layer distance measures.
• To explore graph distance measures for genre-based movie analysis and classification.

Main Methods:

• Constructing multilayer networks from movie scripts (character, keyword, location).
• Applying five graph distance measures: network portrait divergence, NetLSD, NetMF, Laplacian spectra, and D-measure.
• Analyzing movie similarities within and across genres (sci-fi, horror, romance, comedy).

Main Results:

• Different graph distance measures effectively identify similarities among movies within the same genre.
• The methods show varying success in classifying movies based on genre.
• The effectiveness of a method depends on its ability to capture the network's inherent structure.

Conclusions:

• Multilayer network models provide a more comprehensive approach to movie story analysis.
• Graph distance measures are valuable tools for understanding genre-based movie similarities.
• The proposed methodology can be integrated into movie recommendation systems for improved performance.