Using DialoGPT for Text Classification

I have labelled set of 21000 chat transcripts between agent and customer. My intent is to classify what they are talking about? I have already tried few models but wanted to try something that was trained on similar kind of data. I stumbled upon DialoGPT. How can I finetune DialoGPT for Multi-class Text Classification?


You have two ways of doing this.
First is through an intent action chatbot (which looks like it’s something you need instead of a generative model). If you want a better control over the responses your chatbot gives and have a set of answers (actions) to user inputs (intents) you should rather solve this as a sequence classification problem where you have user inputs and associated classes (like intent is “greetings” and user input samples are hello, hi, good morning etc). If you want to keep it simple you can just train a sequence classification model and define answers. You can also look at various chatbot frameworks like rasa OS.

Second way is dialoGPT like model where you have conversation turns and you use it to fine tune DialoGPT (nice tutorial here). Note that your responses will be quite random.