Hi

you can parse response.source_nodes to get the source text chunks + document id's

@jerryjliu0 thanks for your response! I agree that we can use response.source_nodes. But you.com and perpexity.ai add inline citations. How to determine which line is coming from which source document. Should I do text/keyword matching to determine this? We might have false attributions by this approach..

Ahhh I see what you're saying. Don't think we have this feature yet, but can look into adding!

@jerryjliu0 paper-qa based on langchain asks the LLM in the prompt itself to specify the citation. See this prompt: https://github.com/whitead/paper-qa/blob/7173e053767d9102b6685bb6b65eafcf5ebf4a7f/paperqa/qaprompts.py#L19. It has "...For each sentence in your answer, indicate which sources most support it ..".

How trustworthy do you expect such attributions to be? Can you think of any alternative approaches?