NineRec: A Benchmark Dataset Suite for Evaluating Transferable Recommendation

• DOI: 10.48550/arxiv.2309.07705
• Publication Date: 2023-01-01
• Authors: Jiaqi Zhang; Yu Cheng; Yongxin Ni; Yunzhu Pan; Yuan Zheng; Junchen Fu; Youhua Li; Jie Wang; Fajie Yuan
• Keywords: Benchmark (surveying); Transfer of learning; Modality (human–computer interaction)

Learning a recommender system model from an item's raw modality features (such as image, text, audio, etc.), called MoRec, has attracted growing interest recently. One key advantage of MoRec is that it can easily benefit advances in other fields, such natural language processing (NLP) and computer vision (CV). Moreover, naturally supports transfer learning across different systems through features, known transferable systems, or TransRec. However, so far, TransRec made little progress, compared to groundbreaking foundation models the fields NLP CV. The lack large-scale, high-quality recommendation datasets poses major obstacle. To this end, we introduce NineRec, dataset suite includes large-scale source domain nine diverse target datasets. Each item NineRec represented by text description high-resolution cover image. With implement end-to-end training manner instead using pre-extracted invariant features. We conduct benchmark study empirical analysis our findings provide several valuable insights. support further research, make code, datasets, benchmarks, leaderboards publicly available at https://github.com/westlake-repl/NineRec.