Transformative Movie Discovery: Large Language Models for Recommendation and Genre Prediction
Abstract
In the era of digital streaming platforms, personalized movie recommendations, and genre prediction have become pivotal for enhancing user engagement and satisfaction. With the growing number of OTT (Over-The-Top) platforms like Netflix, Amazon Prime Video, and Disney+, the need for effective content recommendation systems has become increasingly important. This paper presents a novel approach to movie discovery leveraging large language models. We explore the transformative potential of these models in two primary aspects: genre prediction and recommendation. We evaluated 4 different LLMs across the MovieLens-100K dataset using different prompting techniques. We employed audio subtitles from movie trailers as prompts for large language models (LLMs). We generated movie summaries from these prompts and subsequently used these generated summaries as additional prompts for predicting movie genres. Also, for the recommendation problem, each user’s interaction history with movies is provided, along with the associated generated summary information. Our experiments reveal that large language models excel in capturing complex user preferences and movie characteristics, leading to highly accurate and personalized movie recommendations. Ranking metrics demonstrate that our approach consistently outperforms existing supervised baseline models in movie recommendation tasks.