At the 2024 Inclusion Bund Conference, Ant Group grandly launched its knowledge-enhanced large model service framework KAG, marking its significant progress in building knowledge-enhanced professional agents. The editor of Downcodes learned that this framework was introduced by Liang Lei, the person in charge of Ant Group’s knowledge graph. It cleverly combines knowledge graphs and large models, aiming to improve the accuracy and logical rigor of decision-making in vertical fields, and effectively solve the practical problems of large language models. Challenges faced in application, such as lack of domain knowledge, unreliable complex decision-making, and insufficient factuality. The launch of the KAG framework has undoubtedly injected new vitality into the application of artificial intelligence in professional fields.
At the Inclusion Bund Conference in 2024, Ant Group shared its latest progress in building knowledge-enhanced professional agents, and launched the research and development results of combining knowledge graphs and large models - the knowledge-enhanced large model service framework KAG.
The framework was introduced by Liang Lei, head of Ant Group’s knowledge graph, and aims to guide decision-making and retrieval through graph logical symbols, significantly improving the accuracy and logical rigor of decision-making in vertical fields.
The KAG framework combines the capabilities of Ant's self-developed graph database TuGraph-DB to provide efficient knowledge storage and retrieval capabilities. It has been applied in Alipay's latest AI native app "Zhi Xiaobao", which has increased the accuracy of government question and answer scenarios to 91%, and the accuracy of vertical indicator interpretation of medical question and answer exceeds 90%.
Liang Lei revealed that the KAG framework will be further opened to the community and provide native support in the open source framework OpenSPG to encourage the community to participate in co-construction. The release of the KAG framework not only demonstrates Ant Group’s technical strength in the field of AI, but also provides the industry with a new solution to address the challenges faced by large language models when applied in vertical fields, such as lack of domain knowledge and complexity. Issues such as unreliability and lack of factuality in decision-making.
The KAG framework improves the synergistic effect of large language models and knowledge graphs through five enhancements, including enhancement of knowledge representation, mutual indexing of graph structure and text, symbol-guided disassembly and reasoning, concept-based knowledge alignment, and KAG Model. This achievement is expected to promote the application of AI in the field of professional services and improve the accuracy and reliability of services.
Project address: https://github.com/OpenSPG/openspg
The open source of the KAG framework will further promote the progress and application of artificial intelligence technology, provide more developers with powerful tools, and jointly promote the development of the AI field. It is believed that in the future, the KAG framework will play an important role in more vertical fields and bring users a more accurate and reliable service experience.