OpenSPG
Can Kag realize character relationship reasoning?
For example, in a report, character A and character B participate in an activity together. When extracting kag, it can establish a "possible_recognize" relationship between character A and character B. Is there a way to achieve such reasoning?
For example, in a report, character A and character B participate in an activity together. When extracting kag, it can establish a "possible_recognize" relationship between character A and character B. Is there a way to achieve such reasoning?
Link prediction is a classic problem in the field of knowledge graph. The goal of KAG-Builder is to build knowledge graph from docs, and then most of (gnn algorithm / pattern matching algorithm ) can be applied in KAG-Solver.
For example, in a report, character A and character B participate in an activity together. When extracting kag, it can establish a "possible_recognize" relationship between character A and character B. Is there a way to achieve such reasoning?
Link prediction is a classic problem in the field of knowledge graph. The goal of KAG-Builder is to build knowledge graph from docs, and then most of (gnn algorithm / pattern matching algorithm ) can be applied in KAG-Solver.
I understand that the solver module is mainly for intelligent QA. Does it mean that the solver module extracts the character reasoning relationship by constructing prompts and then writes it into the graph?
character A and character B participate in an activity together
Extracting potential relationships between characters from the context of an article can indeed compensate for the semantic limitations of vector recall. However, this extraction process is also costly, primarily because extraction and application are two separate stages, making it difficult to predict user questions during extraction.
KAG 0.8 improved the management of private domain knowledge base indexing, incorporating multiple foundational index types such as Outline, Summary, KnowledgeUnit, AtomicQuery, Chunk, and Table. You can try AtomicQuery in your scenario.