RECAP: Feedback-Driven Streaming Semantic User Profiles for Short-Video Recommendation

2026-07-17

AI Summary

This work addresses the limitations of existing open-loop semantic user profiling methods, which struggle to continuously update profiles and directly optimize recommendation performance in streaming short-video scenarios. To overcome this, the authors propose RECAP, a novel framework that introduces a closed-loop feedback mechanism into streaming semantic user profile generation for the first time. RECAP leverages a large language model (LLM) to filter user behaviors and generate semantic summaries, constructing a structured memory with explicit lifecycle and capacity control. A dual-tower evaluator produces GRPO-based reinforcement learning rewards, creating profile-guided semantic feedback signals to refine the profiling policy. Evaluated on a real-world Kuaishou dataset, RECAP achieves a 0.0084 gain in uAUC and a 4.9% improvement in Recall@2000; online A/B tests further demonstrate a statistically significant 0.139% increase in average user watch time per capita.

Abstract

Language-based user profiles convert long behavioral histories into explicit semantic representations for recommendation. However, most profile generators are optimized in an open loop: they may summarize past behavior fluently, but are not directly trained to improve future recommendation. We study this problem in real-world short-video recommendation, where user behaviors continuously arrive as streams and profiles must be incrementally updated under limited capacity. This requires maintaining a consistent bounded profile state and constructing profile-targeted semantic feedback from industrial implicit behavior logs. We propose RECAP, an offline closed-loop framework for optimizing streaming structured semantic profiles with historical recommendation feedback. RECAP maintains each profile as a bounded structured memory by combining LLM-based semantic updates with deterministic lifecycle and capacity control. RECAP constructs profile-targeted semantic feedback by filtering label-consistent behavior pairs with an LLM judge and training a dual-tower evaluator whose matching score serves as a GRPO reward. Experiments on Kuaishou short-video data show that RECAP improves uAUC by 0.0084 and Recall@2000 by about 4.9% over the base generator. Further analyses confirm the benefits of feedback construction and policy optimization, and show more grounded refinement and user-level abstraction in profile updates. A seven-day online A/B test further shows a statistically significant 0.139% improvement in average application usage time per user.

Problem

• user profiling
• short-video recommendation
• streaming data
• semantic feedback
• closed-loop optimization

Innovation

• closed-loop optimization
• streaming semantic profiles
• LLM-based feedback construction
• bounded structured memory
• GRPO reward