Using Elasticsearch, is it possible to query for documents which are not converted into embeddings?

[ajaykrupalk]

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

const vectorstore = new ElasticVectorSearch(embeddings,clientArgs);
// retrieve the document from the vectorstore
const retriever = vectorstore.asRetriever();

Description

I am trying to use Elasticsearch with Langchain and I would like to know how I could get documents which are not converted to embeddings. Only a few of my data is converted into embeddings but most of it doesn't. How do I take this forward?

System Info

NA

[dosubot[bot]]

Hey @ajaykrupalk! Great to see you diving into another interesting challenge with LangChain. How's it going?

To query for documents in Elasticsearch that have not been converted into embeddings when using the LangChain JS framework, you would typically follow these steps:

1. Identify Unconverted Documents: First, you need a way to identify which documents have not been converted into embeddings. This could be done by maintaining a flag or a status field in your Elasticsearch documents that indicates whether a document has been converted.

2. Update Document Status: When a document is converted into embeddings using the LangChain framework, you should update this status in Elasticsearch to indicate that the document has been processed.

3. Elasticsearch Query: To find documents that have not been converted into embeddings, you would query Elasticsearch for documents where the status field indicates they have not been processed.

Given the context does not provide details on how documents are stored in Elasticsearch or if there's a mechanism in place to track their conversion status, a generic Elasticsearch query example to find unconverted documents might look like this:

const { Client } = require('@elastic/elasticsearch');
const client = new Client({ node: 'http://localhost:9200' });

async function findUnconvertedDocuments() {
  const { body } = await client.search({
    index: 'your_index_name',
    body: {
      query: {
        bool: {
          must_not: [
            {
              exists: {
                field: "converted_to_embeddings"
              }
            }
          ]
        }
      }
    }
  });

  console.log(body.hits.hits);
}

findUnconvertedDocuments();

In this example, converted_to_embeddings is a hypothetical field you would add to your documents to indicate their conversion status. The query looks for documents where this field does not exist, implying they have not been converted into embeddings.

Remember, the actual implementation details would depend on how you're managing your documents in Elasticsearch and the mechanism you choose to track whether a document has been converted into embeddings.

As for how LangChain JS handles the conversion of documents into embeddings, it does so through the GoogleVertexAIMultimodalEmbeddings class. This class is designed to generate embeddings for text and images using Google's Vertex AI. The conversion of documents into embeddings involves preparing the media in a format accepted by Google Vertex AI, sending a request to generate embeddings, and then formatting the response into a structured format that includes the embeddings.

Lastly, LangChain JS interacts with Elasticsearch through the ElasticVectorSearch class, which is part of the @langchain/community/vectorstores/elasticsearch module. This class is specifically designed to handle the interaction between LangChain and Elasticsearch, facilitating operations such as adding documents to the Elasticsearch index, performing similarity searches, and deleting documents.

I hope this helps! If you have any more questions, feel free to ask.

Sources

• examples/src/retrievers/embeddings_filter.ts
• langchain/src/experimental/multimodal_embeddings/googlevertexai.ts
• examples/src/indexes/vector_stores/elasticsearch/elasticsearch.ts

About Dosu

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[ajaykrupalk]

@dosu-bot When I set ElasticVectorSearch as my Vector Store, it is only retrieving documents that have been converted to embeddings, while 98% of my data is just some sort of structured or unstructured data that is not converted to embeddings. I would like to retrieve those documents as well and perform some operations

const vectorstore = new ElasticVectorSearch(embeddings,clientArgs);
//retrieve the document from the vectorstore
const retriever = vectorstore.asRetriever();