For example when I searched for Disney

Vector search is best suited for semantic search (i.e. using a phrase to find similar pieces of text), rather than querying a single word

You might be more interested in the keyword index ?

Thank you for your suggestion! Actually I am searching for a full name (surname plus given name), which can also be considered as a phrase? My use case is to make a knowledge base application for customers, so I think semantic search is more suitable. I will try keyword indexing, is there any way to combine different indexes together?

Also, when I search for the phrase 'Please introduce someone', some people don't show up in the results, but some do.

Yea! You could create two indexes (a vector and keyword) and define a tree index on top of those:
https://gpt-index.readthedocs.io/en/latest/how_to/index_structs/composability.html

https://gpt-index.readthedocs.io/en/latest/guides/tutorials/graph.html

Yea, that's kind of the nature of embeddings lol. By default, its returning the top 1 matching node

You can increase the top k nodes using something like index.query(..., similarity_top_k=3, response_mode="compact")

I set similarity_top_k to 4, but there are still no results for some people. 😂 Thank you very much for your answer, I will try to combine index!