Tencent's VerRec uses verbalized feedback instead of scalar rewards for generative recommendation RL
_reachsumit · x · 2026-11-18
Tencent researchers propose VerRec, an RL framework for generative recommendation that replaces scalar rewards with personalized verbalized feedback from a user simulator.
- The simulator critiques each candidate in words, letting the policy know what conflicts with user preferences and how the next candidate should differ
- Original and refined candidates form expanded groups jointly optimized with GRPO; the simulator is only needed at training time
- Experiments on three Amazon review datasets and two generative backbones show improved accuracy and ranking quality, with verbalized guidance encouraging more diverse exploration
More from Research
- Voyager 1 Reaches One Light-Day From Earth, Signals Now Take 24 Hours — snowclipsed · 2026-11-18
- TCS Hiring Statement Urges Quality Over Quantity as AI Fuels Paper Arms Race — fortnow · 2026-11-18
- New paper: a sociocultural approach to linguistic creativity — abenitezburraco · 2026-11-18
- 27B Model Rewrites Its Own Weights Without Training or Gradients, Paper Claims — udmrzn · 2026-11-18
- From 'impossible' to a working Bluetooth beacon: open-source EDA chip design is real — IgorCarron · 2026-11-18
- AURA 2.0: a streaming video LLM that speaks first, calls tools, and grounds visually — Shuhan Li · 2026-11-18