For NLP intent analysis, which is more effective: using a custom dataset or leveraging an LLM model?
Hi @yogeswaranc !
So with intent analysis you typically want a finite set of states (“intents”), regardless of whether you go for a bespoke or a fine-tuned model, or a generalist LLM. And with both cases you need to have sufficient samples fairly uniformly distributed for each state/intent.
With a generalist model, like GPT-4o, you could get away with much fewer set of examples for each intent. So then you would employ “few shot prompting”. If you are thinking of training and deploying your own intent classification model (e.g. not using fine-tuning API, but actually training your own model), then you have even more complexity in terms of setting up, training, and deploying and managing your model.
So the answer is highly dependent on your situation - but if you don’t have a lot of data, or if you don’t have lot of expertise in training your own model, then I would first try out few-shot prompting technique using e.g. GPT-4o.