DualGR: Generative Retrieval with Long and Short-Term Interests Modeling

2025-11-16

AI Summary

Large industrial recommender systems face three key challenges in generative retrieval (GR): coupling of long-term and short-term user interests, high noise in semantic ID (SID) generation, and insufficient modeling of negative feedback from unclicked exposed items. To address these, we propose DualGR—a dual-branch generative retrieval framework. Its core contributions are: (1) a Dual-Branch Router (DBR) that explicitly disentangles long-term user preferences from short-term intent; (2) Search-enhanced Hierarchical SID Decoding (S2D), which mitigates contextual noise via search-guided hierarchical decoding; and (3) Exposure-aware Next-item Prediction Loss (ENTP-Loss), treating unclicked exposed items as hard negatives. DualGR further incorporates cross-attention to model behavioral sequences. Deployed on Kuaishou’s short-video recommendation system, A/B testing demonstrates statistically significant improvements: +0.527% in video play rate and +0.432% in average watch time—validating DualGR’s dual advantages in relevance and response efficiency.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Natural Language Processing: Generation
• Machine Learning: Learning Preferences or Rankings

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

In large-scale industrial recommendation systems, retrieval must produce high-quality candidates from massive corpora under strict latency. Recently, Generative Retrieval (GR) has emerged as a viable alternative to Embedding-Based Retrieval (EBR), which quantizes items into a finite token space and decodes candidates autoregressively, providing a scalable path that explicitly models target-history interactions via cross-attention. However, three challenges persist: 1) how to balance users' long-term and short-term interests , 2) noise interference when generating hierarchical semantic IDs (SIDs), 3) the absence of explicit modeling for negative feedback such as exposed items without clicks. To address these challenges, we propose DualGR, a generative retrieval framework that explicitly models dual horizons of user interests with selective activation. Specifically, DualGR utilizes Dual-Branch Long/Short-Term Router (DBR) to cover both stable preferences and transient intents by explicitly modeling users' long- and short-term behaviors. Meanwhile, Search-based SID Decoding (S2D) is presented to control context-induced noise and enhance computational efficiency by constraining candidate interactions to the current coarse (level-1) bucket during fine-grained (level-2/3) SID prediction. % also reinforcing intra-class consistency. Finally, we propose an Exposure-aware Next-Token Prediction Loss (ENTP-Loss) that treats "exposed-but-unclicked" items as hard negatives at level-1, enabling timely interest fade-out. On the large-scale Kuaishou short-video recommendation system, DualGR has achieved outstanding performance. Online A/B testing shows +0.527% video views and +0.432% watch time lifts, validating DualGR as a practical and effective paradigm for industrial generative retrieval.

Problem

• Modeling long and short-term user interests in generative retrieval
• Reducing noise in hierarchical semantic ID generation
• Incorporating negative feedback from exposed but unclicked items

Innovation

• Models dual user interests with selective activation
• Controls noise via search-based hierarchical ID decoding
• Incorporates exposure-aware negative feedback in training