一种基于大语言模型的人机交互行为预测方法及系统
By combining large-scale language models and hybrid retrieval techniques, multimodal instructions are parsed and a local knowledge base is built. Semantic tags are integrated for vector and graph retrieval, and standardized knowledge fragments are generated. This solves the limitations of local properties and data security issues in human-computer interaction systems, and achieves efficient and accurate behavior prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIAODUO INTELLIGENT TECH (BEIJING) CO LTD
- Filing Date
- 2025-11-04
- Publication Date
- 2026-07-17
AI Technical Summary
Existing human-computer interaction systems suffer from limitations in locality and data security when processing domain-specific knowledge. Furthermore, large language models may produce illusions when generating responses, making it difficult to make efficient and accurate behavior predictions in unknown or changing situations.
By combining the reasoning capabilities of large-scale language models with hybrid retrieval techniques, a local knowledge base is constructed by parsing multimodal instructions, integrating instruction and scene semantic tags, performing vector and graph retrieval, generating standardized knowledge fragments, and optimizing input data using prompt words, ultimately generating accurate behavior prediction results.
It significantly improves the intelligence level and response efficiency of human-computer interaction, enhances the ability to process complex instructions, prevents data leakage and hallucinations, and adapts to changes in unknown situations.
Smart Images
Figure CN121502010B_ABST