Topic modelling with large language models

Did anyone try to extract bigrams, trigrams from production support tickets using llm? I wanted to know if there are any limitations/challenges .

Haven’t tried it specifically for prod support tickets but used them on normal sentences.

A good prompt and a few shot samples should do the trick for you. Did not encounter a lot of challenges per se, though the behaviour can be a bit unpredictable using 0-shot or 1 shot prompts

1 Like

Counting them could be challenging. Maybe splitting the context in chunks and doing multiple evaluations helps you as well.

1 Like