How does vector similarity work I have a

The community member is asking why an exact sentence they sent is not found in the returned nodes, even though the sentence is present in one of the documents. A community member explains that the embeddings generated from a single sentence are different from the embeddings generated for an entire chunk or node, as the node has more context that biases the embeddings. Therefore, searching for exact sentences is not a good way to test vector similarity, as it is more about the similarities in overall semantics. The community member suggests augmenting the vector search by also using a keyword index, and provides an example of setting up a custom retriever to do this.