Binge Watch: Reproducible Multimodal Benchmarks Datasets for Large-Scale Movie Recommendation on MovieLens-10M and 20M
New datasets enrich MovieLens with plots, posters, and trailers using state-of-the-art encoders.
Deep Dive
A team of six researchers led by Giuseppe Spillo has released M3L-10M and M3L-20M, two large-scale, reproducible multimodal datasets. They enrich the classic MovieLens-10M and 20M datasets with movie plots, posters, and trailers, extracting textual, visual, acoustic, and video features using modern encoders. This provides a standardized benchmark for developing and testing advanced Multimodal Recommender Systems (MRSs) with rich side information.
Why It Matters
Provides a crucial, reproducible foundation for building next-generation AI that recommends content using video, audio, and text, not just ratings.