Method and system for automatically generating a video based on unstructured knowledge using multi-agent
By using a multi-agent system to evaluate data source weights and calculate semantic similarity of unstructured knowledge materials, controversial knowledge units in the video generation process are identified and processed. This solves the problems of knowledge authenticity identification and content quality control in automatic video generation, and achieves high-quality video output.
Patent Information
- Authority / Receiving Office
- CN ยท China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- KEBAIWEN (SHENZHEN) TECH CO LTD
- Filing Date
- 2025-12-02
- Publication Date
- 2026-07-21
AI Technical Summary
Existing automatic video generation technologies lack mechanisms for assessing the authenticity and reliability of knowledge, which may result in generated video content containing incorrect or false information, affecting the video's authority and credibility, and making it difficult to achieve intelligent matching between knowledge credibility assessment and video content presentation.
A multi-agent system is used to evaluate the data source weights of unstructured knowledge materials. Repeated or contradictory knowledge fragments are identified through time verification and content semantic similarity calculation. A list of disputed knowledge units is generated, and the semantic vector similarity algorithm is used to analyze the matching degree between the disputed units and the preset video script framework. Combined with the backup knowledge base replacement algorithm, a high-quality video output is generated.
It achieves end-to-end quality control from knowledge identification to video generation, improving the accuracy and credibility of multi-source knowledge fusion videos and ensuring the authenticity and authority of video content.
Smart Images

Figure CN121418636B_ABST