Film Genre Prediction and Analysis Using Multimodal Embeddings from Large Models: The Case of Noir

Le résumé fourni par la source

Abstract: We used embeddings (numerical representations) from open-sourced language and image models to compare and contrast films in the "noir" and "neo-noir" genres with each other and with those in other genres, using as proxy sources the available plot summaries for text and posters as images. Noir as a genre was found to be more distinctive, and therefore predictable through machine learning, than neo-noir, which was more easily confused with other genres. The process of attempted classification itself reveals interesting similarities across genres (e.g., screwball comedy murder mystery and classic noir) as well as the shortcomings of the sources used, which fail to capture less tangible aspects of noir. Overall, we demonstrate the value of these approaches for genre analytics and potential incorporation into discovery tools.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.