Evaluating Multimodal Narrative Understanding of Popular Hollywood Films

Research is underway to improve multimodal language models for large-scale computational analysis of film, aiming to advance understanding of film history and narrative evolution. A primary challenge identified is the copyright protection of Hollywood films, which complicates the creation of stable benchmarks. To address this, a new collection of films is being built, focused on box office popularity and likely public domain status, leveraging newly released historical box office data and copyright research.

Why it matters

The development of robust computational tools for film analysis can unlock new avenues for understanding media trends and cultural evolution, providing deeper insights for strategic decision-making in content creation and preservation. Overcoming data access limitations through legal and historical research is critical for advancing this technological capability across broader media datasets.

Key insights

• Multimodal language models offer significant potential for scalable computational analysis of films.
• This analysis can provide insights into film history and the development of narrative techniques.
• Copyright protections on Hollywood films pose a significant hurdle for creating standardized research benchmarks.
• A new film collection is being developed, based on two criteria: box office popularity and probable public domain status.
• The project includes the first large-scale, open collection of weekly box office earnings from Variety magazine (1922-1979).
• Copyright registrations and renewals in the US Catalog of Copyright Entries are being researched to determine public domain status.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.21430