Intelligent question-answering method and device based on tool matching, equipment, medium and program product
By preprocessing long videos and selecting appropriate tools, dynamically adjusting the sampling interval, filtering suitable content parsing tools, extracting fine-grained information, and performing large language model inference, the problem of inaccurate answers to complex questions in long videos has been solved, thereby improving the accuracy and effectiveness of intelligent question answering.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA MOBILE JIUTIAN ARTIFICIAL INTELLIGENCE TECHNOLOGY (BEIJING) CO LTD
- Filing Date
- 2026-03-05
- Publication Date
- 2026-06-05
AI Technical Summary
Traditional end-to-end models struggle to extract fine-grained features from long videos, leading to inaccurate and unreliable answers to complex user questions.
By receiving long video files and user questions, preprocessing is performed, the sampling interval is dynamically adjusted for frame sampling and compression, and matching content parsing tools such as automatic speech recognition, text recognition, face recognition and target tracking tools are selected to extract fine-grained key information, and a large language model is used for reasoning to generate answers.
It improves the accuracy and reliability of intelligent question answering in long video scenarios, makes up for the lack of targeted information extraction in traditional models, and achieves improved accuracy and effectiveness.
Smart Images

Figure CN122157106A_ABST