[Question]: Support for Knowledge graphs

infiniflow / infinity

The AI-native database built for LLM applications, providing incredibly fast hybrid search of dense vector, sparse vector, tensor (multi-vector), and full-text

https://infiniflow.org

Apache License 2.0

2.47k stars 267 forks source link

[Question]: Support for Knowledge graphs #1060

Open acrigney opened 5 months ago

acrigney commented 5 months ago

Describe your problem

Guys I would like to encorporate knowledge graphs into the semantic searches. I see now that neo4j is supporting this, also DGraph and others, although again they generally use Python interfaces that I am not a fan of. Here is an example of how neo4j does it. https://towardsdatascience.com/efficient-semantic-search-over-unstructured-text-in-neo4j-8179ad7ff451 Does infinity support this already? Or could this be implemented. Really my experience with semantic search is that they benefit from improved context for the searches.

yingfeng commented 5 months ago

This could be implemented later. However, graph search is weaker than keyword demand, regarding to knowledge graph, only single order semantic is required for most cases, which could also be provided through full text search.

acrigney commented 5 months ago

Wow mate! So far with my experience simple semantic search has not been good enough for my customers, but for my graph based semantic search with neo4j I think I just need to improve my prompts. However there must have been some upgrades in Python that broke their code and as Python seems to have such poor dependency handling its really hard to work out whats wrong. Whearas if I build everything with C# with nuget packages I will know everything in detail if anything goes wrong. Hence I want to build this reference RAG stack all in C#.