How can I get document id when using PGVector similaritySearch

[Exlau]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

-

Description

When I'm trying to use "PGVectorStore.similaritySearch(query, limit)", the results are like this [{metadata:{...},pageContent:""}], I want to get the document id also, how can I do that?

System Info

node-ver: 16.20.1

[dosubot[bot]]

Hey @Exlau! I'm here to assist you with any bugs, questions, or contributions you have. Let's tackle this together!

To get the document ID when using PGVectorStore.similaritySearch(query, limit), you should use the similaritySearchVectorWithScore method. This method returns an array of tuples, each containing a Document and its similarity score. The Document object includes the document ID.

Here is an example of how you can use this method:

const results = await pgVectorStore.similaritySearchVectorWithScore(query, limit);

results.forEach(([document, score]) => {
  console.log(`Document ID: ${document.id}, Score: ${score}`);
});

In this example, pgVectorStore is an instance of PGVectorStore, query is your query vector, and limit is the number of most similar documents you want to retrieve. The document.id will give you the document ID ^[1].

To continue talking to Dosu, mention @dosu.

[MJDeligan]

Since similarity search returns only documents and documents do not include their id, there is currently no way to get the id when doing similarity search. If you insert with a custom id, you could also set that as an id-field in the metadata and retrieve it that way. Otherwise I'm afraid there is currently no other way except overwriting the PGVectorStore to include the id in the return. There's been some discussion about adding the id to the document interface and @jacoblee93 has mentioned that a new type of vectorstore/indexing integration is under works or at least being considered.

[Exlau]

Good to know that, btw is there any ongoing branch or PR?