Generative Recommendation
Overview
Generative recommendation studies how to produce personalized items or ranked lists directly from user histories and context, instead of scoring a fixed candidate set. Common approaches encode items as semantic IDs and train autoregressive Transformers or language models, while diffusion, flow-based generation, and continuous retrieval offer alternatives. Training combines next-item supervision with semantic or behavioral signals, and sometimes reinforcement learning; evaluation uses sequential benchmarks such as Amazon datasets alongside production A/B tests. A central challenge is balancing useful item representations and reachability—including for cold or long-tail items—with accuracy, fairness, and the latency demands of large-scale serving.
Latest papers
107 papers11 papers in the last four weeks, down 15% on the four weeks before. 0.1% of all new papers.