How to make Retrieval QA return the exact documents used for answering the query

[pi-crust]

I'm creating a QA bot with RAG and aiming to provide the specific documents from which the answers are extracted. While Retrieval QA retrieves all source documents via the retriever, it uses only a subset to answer a question. How can I extract these specific documents from Retrieval QA? Is there a method for achieving this?

[dosubot[bot]]

🤖

Hello @pi-curst! 👋
I'm Dosu, a bot here to assist you with your issues, questions, and contributions to the langchainjs repository while we wait for a human maintainer. Let's work together to get things rolling!

To achieve your goal of having the Retrieval QA return the exact documents used for answering a query, you can utilize the returnSourceDocuments option in the RetrievalQAChain class. This option, when set to true, will include the source documents used to answer the query in the response.

Here's how you can set it:

const chain = RetrievalQAChain.fromLLM(model, vectorStore.asRetriever(), {
  returnSourceDocuments: true, // Can also be passed into the constructor
});

In the response from the chain, the sourceDocuments field will contain an array of the documents used to answer the query. Each document in the array will have a pageContent field containing the text of the document and a metadata field containing any additional information about the document.

You can find more details in the examples/src/chains/retrieval_qa_sources_legacy.ts file.

Also, it's worth noting that the returnSourceDocuments option is false by default in the RetrievalQAChain class. However, it can be overridden during the instantiation of the class. You can find this in the langchain/src/chains/retrieval_qa.ts file.

I hope this helps! If you have any other questions or need further clarification, feel free to ask.

Sources

• examples/src/chains/retrieval_qa_sources_legacy.ts
• langchain/src/chains/retrieval_qa.ts

---

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.

[pi-crust]

How do we know which of those source documents that LLM extracted the answer from?

Retrieval QA uses k documents which are semantically similar to generate the answer. The answer need not be in all the k documents, how do we know which documents out of the k documents the answer is extracted from?

[jdjayakaran]

Was there a solution found for the same?
I have the same error too