How to work with a low quality knowledge base

Hey everyone,
I am trying to build a bot that would answer users questions using a knowledge base of our resolved support tickets.
The goal would be that the bot is able to resolve recurring problems that have already been solve before.

My problem is that my knowledge base has a lot of irrelevant tickets, for example with no explanations of how it was solved or with a badly exposed/unclear problem.

I was thinking of creating an AI Task that would try to clean up the knowledge base, but from what I can find the AI task canno’t directly interact with it ?

For now, I use the knowledge base description to influence the bot to only use the relevant informations he can find.
But I wonder if there are better solution to this ? (Beside cleaning up the dataset by hand)

Thanks !

Hi @hugo.delautre ,

From a conversational design point of view, the best thing is to direct the user through a flow where you can narrow their problem then from there; you can let the KB answer directly.

The other thing is, can you clean the tickets and keep the relevant ones?

1 Like

Thank you for your help.

As for cleaning the tickets, I guess it is theoreticaly possible, but would take a lot of time and I’d prefer it to be the last solution.

I will try implementing some some flow to narrow it down and see how it goes.

Thanks agin !

Your welcome, keep us posted about what worked the best for you :slight_smile: