[Question]: When uploading files in a chat, should the chat scope be narrowed down?

[kuschzzp]

Describe your problem

I upload files while chatting, and when I ask questions about what's in the files, the answers aren't accurate.

My idea of this phenomenon is that if I upload files, then the scope of my conversation should be limited to the current file, rather than the entire knowledge base.

What do you think of this?

[KevinHuSh]

Good suggestion.

[alliecatowo]

Maybe there could be a config option to assign a weight to more recent files, or weight the chat files heavier than the knowledge base? So between 2 files with the same similarity score, one from the knowledge base, and one from the chat history, it would prefer the chat file.

A ui element to toggle retrieval from the knowledgebase could be a nice and elegant solution as well.

[kuschzzp]

@KevinHuSh @alliecatowo Thanks for reply.

The current performance of the main branch code is:

• Create A knowledge base and upload a file.
• Create dialog and conversation to ask questions about the content in A. 【 Can get the correct answer 】
• Upload a file B in conversation and ask questions in B. 【 Get the right answer 】
• Upload a file C in conversation and ask questions in C. 【 Get the right answer 】
• Direct questions to B without attachments. 【Can get correct reply】
• Direct questions to C without attachments. 【Can get correct reply】
• In this case, ask questions in file A in the knowledge base without attachments. 【 Did not get a correct reply 】

Once the file is uploaded, the original knowledge in the knowledge base is lost.

---

I think the better phenomenon is this:

When I include attachments in this conversation, the answer should be based only on this attachments.

When I do not include attachments in this conversation, the answer should be based on the contents of the knowledge base and the contents of the attachments previously uploaded.

---

To implement the phenomenon I mentioned, you simply need to remove the following code：

ragflow/api/db/services/dialog_service.py

Lines 207 to 209 in 754d5ea

	for m in messages[:-1]:
	if "doc_ids" in m:
	attachments.extend(m["doc_ids"])