The post asks if the semanticsplitternodeparser works in a certain way. One community member confirms that it "basically works like that". Another community member has a follow-up question about document embeddings - they ask what happens to the whole-document vectors when a document is split into chunks and each chunk gets its own vector. The response indicates that there are no whole-document embeddings, only node embeddings, and the terms "node" and "document" are used interchangeably.
I have a question about this. 'Document Embeddings' creates a vector for each document, but then a document is split into chunks and a vector is created for each chunk, but what happens to the whole-document vectors? Are they retained along side the chunk-vectors? Is some parent-child relation maintained or does that have to be described with metadata or something?
