I finally got my AI to stop mixing up similar product names

I was building a simple chatbot for a small online store in Portland, and it kept confusing 'blueberry jam' with 'blackberry jam' in customer orders. After two weeks of tweaking the training data with no luck, I tried adding just five specific example sentences for each product, like 'I want the blueberry jam on toast'. It fixed about 90% of the errors overnight. Has anyone else found that a tiny bit of very clear data works better than a huge, messy dataset for these kinds of problems?

3 comments

3 Comments

olivia_chen351mo ago

I always thought more data was better, but your fix makes total sense now!

susanh4612d ago

Honestly, my buddy's whole project crashed because he overloaded it with useless info.

pipera501mo ago

Yeah, that was my old mindset too (until I saw the mess it made in practice).