Hi, I am feeling lost between when to use tree index and when to use vector one. I understand how they work, but I do not fully understand why it would make more sense to use vector index with a manual and not a tree one. Semantic search for vector would help me identify the most likely selection for LLM, but with tree index It would give me an ability to find the most relevant answer by drilling down.
This is where I really do not understand the reasoning. Vector index is based on words and proximity of other words to it. By using semantic search we are finding finding alternative words in the question that may be in vector index. This seems to be it. So, depending on the semantics, there is a really good chance that it will choose a different node.
If there are 10 pages each talking about how to use one or another online form, the only chance it has of getting the right page is if there is a unique word that would distinguish it from other pages and during the embedding it will deem that word important enough... I do not see how it would correctly choose the right node .