While running ChatBot with data from the Fediverse, I noticed that it would often select matches that have a lower confidence if the minimum confidence was not reached. Would it make sense to select the result with the highest confidence instead?

One issue I noticed is that it would sometimes end up picking matches that lack any responses, and then would fall back to picking a random response. The end-user experience was similar in either case: a seemingly random response.

1. Without this set of changes, when a match with a low confidence was chosen and a response was found.
2. With this set of changes, when a match with a high confidence—but with no responses—was chosen, which resulted in a random response being chosen instead.