3 ms·
Generative Knowledge with Semantic Graphs, LLMs and RAG
- dmezzetti 3y agoSemantic graphs use vector models to build a graph network with semantically similar relationships generated for a dataset. While this can be used standalone, it also pairs well with RAG. This differs from a standard vector query for RAG in that this method not only looks at semantic similarity between the search and results but the similarity within the results. It's effectively judging the results to see which result is most similar to other results. Graph analysis algorithms such as page rank and centrality can be used to form LLM contexts for RAG. This method is best for knowledge discovery and topic exploration when there isn't a clear question to ask. Sometimes the hardest thing about search is knowing what to search for and that's the main purpose here.