A collection of resources and papers of Multimodal Recommender Systems (MRS).
🔥🔥 We will update the repo sustainably!
In our survey, we conclude the general MRS as an unified process, including Raw Feature Representation, Feature Interaction and Recommend Model. To face with the challenges contained in each procedure, we classify the existing works according to four branch of techniques, i.e., Modality Encoder, Feature Interaction, Feature Enhancement and Optimization.
More details can be seen in our survey.
There are two open-sourced repositories for implementing multimodal recommender system models.
MMRec: A PyTorch benchmark, which implements 15 most recent MRS models.
Cornec: A PyTorch framework, which implements more earlier MRS model.
Data | Field | Modality | Scale | link |
---|---|---|---|---|
Tiktok | Micro-video | V,T,M,A | 726K+ | link |
Kwai | Micro-video | V,T,M | 1M+ | link |
Movielens+IMDB | Movie | V,T | 100K-25M | link |
Douban | Movie, Book, Music | V,T | 1M+ | link |
Yelp | POI | V,T,POI | 1M+ | link |
Amazon | E-commerce | V,T | 100M+ | link |
Book-Crossings | Book | V,T | 1M+ | link |
Amazon Books | Book | V,T | 3M | link |
Amazon Fashion | Fashion | V,T | 1M | link |
POG | Fashion | V,T | 1M+ | link |
TMall | Fashion | V,T | 8M+ | link |
Taobao | Fashion | V,T | 1M+ | link |
Tianchi News | News | T | 3M+ | link |
MIND | News | V,T | 15M+ | link |
Last.FM | Music | V,T,A | 186K+ | link |
MSD | Music | T,A | 48M+ | link |
Note: ’V’, ’T’, ’M’, ’A’ indicate the visual data, textual data, video data and acoustic data, respectively