#community-help

Limiting Irrelevant Responses in Q&A System

TLDR Gio asked how to limit irrelevant answers, and Kishore Nallan suggested using a threshold on vector_distance value to determine similarity.

Powered by Struct AI
Feb 28, 2023 (9 months ago)
Gio
Photo of md5-fca9a1d3ceee0b8b4bedc936ea93e48b
Gio
12:21 PM
Would you shed some light on this one too? How would I limit such cases?

My collection consists of 3 paragraphs off of Wikipedia about the war in Ukraine.
Obviously it won’t have an answer about how old I am. I’d love to return something like “it is an irrelevant question” or “I couldn’t find an answer”…
Image 1 for Would you shed some light on this one too? How would I limit such cases?

My collection consists of 3 paragraphs off of Wikipedia about the war in Ukraine.
Obviously it won’t have an answer about how old I am. I’d love to return something like “it is an irrelevant question” or “I couldn’t find an answer”…
Kishore Nallan
Photo of md5-4e872368b2b2668460205b409e95c2ea
Kishore Nallan
12:24 PM
You can look at the vector_distance value in the response and have some threshold for determining if it's way out of similarity domain.
Mar 03, 2023 (9 months ago)
Gio
Photo of md5-fca9a1d3ceee0b8b4bedc936ea93e48b
Gio
09:22 PM
Kishore Nallan how much would be ‘way out of similarity’?
Mar 04, 2023 (9 months ago)
Kishore Nallan
Photo of md5-4e872368b2b2668460205b409e95c2ea
Kishore Nallan
02:44 AM
The distance value returned is between 0 to 2. Higher the distance, less similar. The actual threshold will depend on your dataset.