一种面向生成式推荐系统的位置感知投机解码加速方法
By introducing item location encoding and speculative step location encoding into the generative recommendation system, the structural and depth perception capabilities of the draft model are enhanced, the shortcomings of existing speculative decoding techniques are addressed, and a dual improvement in recommendation quality and speed is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- UNIV OF SCI & TECH OF CHINA
- Filing Date
- 2026-01-21
- Publication Date
- 2026-07-17
AI Technical Summary
Existing speculative decoding techniques suffer from insufficient structure awareness and poor depth adaptability in generative recommendation systems, resulting in deviations between the generated candidate sequences and the target distribution, low verification pass rates, and impacted reasoning efficiency.
By introducing item location encoding and speculative step location encoding, and fusing them with the original semantic embedding vector through a gating mechanism, the draft model's ability to perceive item structure and speculative depth is enhanced, thereby improving the generation quality and validation pass rate of candidate sequences.
Without increasing computational overhead, it significantly improves the inference speed and quality of the recommendation system, enabling efficient personalized recommendation services.
Smart Images
Figure CN121860064B_ABST