At the Inclusion·Bund Conference in 2024, Ant Group shared its latest advancements in constructing knowledge-enhanced professional AI agents and introduced the Knowledge-Augmented Large Model Service Framework (KAG), a research achievement combining knowledge graphs with large models.
This framework, presented by Liang Lei, head of Ant Group's knowledge graph, aims to significantly enhance the accuracy and logical rigor of decision-making in vertical fields through graph logic symbol-guided decision-making and retrieval.
The KAG framework integrates the capabilities of Ant's self-developed graph database TuGraph-DB, providing efficient knowledge storage and retrieval capabilities. It has been applied in Alipay's latest AI-native app "ZhiXiaoBao," improving the accuracy rate of government service inquiries to 91% and surpassing 90% in medical inquiry vertical metric interpretation accuracy.
Liang Lei revealed that the KAG framework will be further opened to the community and natively supported in the open-source framework OpenSPG, encouraging community participation in co-construction. The release of the KAG framework not only showcases Ant Group's technological prowess in AI but also offers a new solution for the industry to address challenges faced by large language models in vertical applications, such as the lack of domain knowledge, unreliability in complex decision-making, and insufficient factual accuracy.
The KAG framework enhances large language models and knowledge graphs through five aspects: enhanced knowledge representation, mutual indexing of graph structures and text, symbol-guided decomposition and reasoning, knowledge alignment based on concepts, and the KAG Model. This achievement is expected to promote the application of AI in professional service fields, improving service accuracy and reliability.
Project Address:https://github.com/OpenSPG/openspg