Find answers from the community

Updated 5 months ago

Hello guys, I have a question about Embedding from TextNode with metadata

At a glance

The community members are discussing whether the presence of metadata affects the embeddings when saving a TextNode in a VectorStoreIndex. The original poster noticed that the embeddings were different when creating TextNodes with and without metadata. The comments indicate that the metadata is stored with the nodes and is also embedded, allowing for similarity searches on the metadata fields. However, the community members also note that specific metadata attributes can be excluded from the embedding model if the Llamaindex documents are created manually.

Useful resources
Hello guys, I have a question: when saving a TextNode in a VectorStoreIndex, does the presence of a metadata be taken into account? I tried to create based on TextNode without and with metadata, their embeddings were different
V
T
r
7 comments
Hello guys, I have a question: when saving a TextNode in a VectorStoreIndex, does the presence of a metadata be taken into account? I tried to create based on TextNode without and with metadata, their embeddings were different
The metadata is stored with the nodes
I understand this, but embedding should be generated only according text field?
The metadata is also embedded, you can do similarity search on the metadata fields
Although the metadata is embedded with the text in a node, you can exclude specific metadata attribute from being passed as input to the embedding model if you created the Llamaindex documents manually
Add a reply
Sign up and join the conversation on Discord