Predicting Movie Hits Before They Happen with LLMs

TL;DR

This work tackles the cold-start problem in movie popularity prediction by using Large Language Models (LLMs) to assess new titles from metadata, delivering both a hit probability and transparent reasoning. The authors build a benchmark dataset from a large entertainment platform and compare LLM-based predictions against embedding-based baselines, using a suite of prompts and model variants (notably Llama-3) to analyze performance. Key findings show that larger LLMs with informative prompts outperform baselines, with listwise ranking generally superior to pairwise in this setting, and that parsing reliability and model-specific prompt design are critical. The study demonstrates practical value for editorial workflows and automated retrieval, offering scalable, metadata-driven screening that can be deployed before extensive interaction data is available, with potential applicability to other domains.

Abstract

Addressing the cold-start issue in content recommendation remains a critical ongoing challenge. In this work, we focus on tackling the cold-start problem for movies on a large entertainment platform. Our primary goal is to forecast the popularity of cold-start movies using Large Language Models (LLMs) leveraging movie metadata. This method could be integrated into retrieval systems within the personalization pipeline or could be adopted as a tool for editorial teams to ensure fair promotion of potentially overlooked movies that may be missed by traditional or algorithmic solutions. Our study validates the effectiveness of this approach compared to established baselines and those we developed.