OneSearch-V2: The Latent Reasoning Enhanced Self-distillation Generative Search Framework
Summary: arXiv:2603.24422v1 Announce Type: cross
Abstract
Generative Retrieval (GR) has emerged as a promising paradigm for modern search systems. Compared to multi-stage cascaded architecture, it offers advantages such as end-to-end joint optimization and high computational efficiency. OneSearch, as a representative industrial-scale deployed generative search framework, has brought significant commercial and operational benefits. However, its inadequate understanding of complex queries, inefficient exploitation of latent user intents, and overfitting to narrow historical preferences have limited its further performance improvement.
To address these challenges, we propose OneSearch-V2, a latent reasoning enhanced self-distillation generative search framework. It contains three key innovations:
- Thought-augmented complex query understanding module: This innovation enables deep query understanding and overcomes the shallow semantic matching limitations of direct inference.
- Reasoning-internalized self-distillation training pipeline: This pipeline uncovers users’ potential yet precise e-commerce intentions beyond log-fitting through implicit in-context learning.
- Behavior preference alignment optimization system: This system mitigates reward hacking arising from the single conversion metric, addressing personal preference via direct user feedback.
Performance Evaluation
Extensive offline evaluations demonstrate OneSearch-V2’s strong query recognition and user profiling capabilities. Online A/B tests further validate its business effectiveness, yielding:
- +3.98% item Click-Through Rate (CTR)
- +3.05% buyer conversion rate
- +2.11% order volume
Manual evaluation further confirms gains in search experience quality, highlighting improvements such as:
- +1.65% in page good rate
- +1.37% in query-item relevance
Addressing Common Search System Issues
More importantly, OneSearch-V2 effectively mitigates common search system issues such as information bubbles and long-tail sparsity. It achieves these improvements without incurring additional inference costs or serving latency, making it a robust solution for modern search challenges.
Conclusion
In summary, OneSearch-V2 represents a significant advancement in generative search frameworks, enhancing the ability to understand complex queries and user intents while improving overall search experience and efficiency. Its innovative design and proven results set a new standard for the future of search systems.