Problem with near_text Query and Metadata Filtering in Weaviate

Hi!

Great! Glad to hear we are making progress

Now, you need to understand that in your scenario, Weaviate will have two indices:

1 - A Named Vector called transcription_vector
2 - An inverted index with all the tokenized content that is searchable and filterable.

So no need to add a second NamedVector only for chave_unica, the filtering happens on the inverted index, and not on the vector index.

You near_text with that filter should work. One thing you can try, but it shouldn’t make a difference, is to set the tokenization to field for the chave_unica property, like so

...
wc.Property(name="chave_unica", data_type=wc.DataType.TEXT, tokenization=wc.Tokenization.FIELD),
...

this will ensure that all content that you set for chave_unica will be considered as a token.

But I am finding it strange that it should work already.

Let me know if you can share the dataset so I can try reproducing it in my end.

THanks!