Adding different search options for the PgVectorKeywordRetriever

[raqkaaq]

Is your feature request related to a problem? Please describe.
The PgVectorKeywordRetriever currently only supports the plainto_tsquery function. This means that any keyword inputs are combined via AND. i.e. "Can cats fly?" -> "cats & fly" in the actual search. There is no method using this function to allow for more nuanced keyword searches. (What if I wanted "cats | fly").

Describe the solution you'd like
Would it be possible to add functionality to support the websearch_to_tsquery, and to_tsquery functions?

Describe alternatives you've considered
There doesn't seem to be alternatives without building a custom keyword retrieval and using metadata to store values. I understand there might be reasons why this hasn't been implemented (security and sql injections), but I am hoping that there is a way to avoid those issues while still giving a more developer friendly way to customize the searches directly from a query.

[anakin87]

Hello!

Our Pgvector Document Store supports some filtering mechanisms that might cover your use case.

I tried the following:

from haystack_integrations.document_stores.pgvector import PgvectorDocumentStore
import os
from haystack import Document

os.environ["PG_CONN_STR"]="postgresql://postgres:postgres@localhost:5432/postgres"
document_store = PgvectorDocumentStore()

print(document_store.count_documents())

docs = [Document(content="I like cats"),
        Document(content="I like dogs"),
        Document(content="Airplanes can fly")]

document_store.write_documents(docs)

filters = {
    "operator": "OR",
    "conditions": [
        {"field": "content", "operator": "like", "value": "%cats%"},
        {"field": "content", "operator": "like", "value": "%fly%"},
    ]
}

print(document_store.filter_documents(filters))
# [Document(id=..., content: 'I like cats'), Document(id=..., content: 'Airplanes can fly')]

Let me know what you think...