# Split long text into "topics"

**URL:** <https://discuss.huggingface.co/t/split-long-text-into-topics/33896>\
**Category:** 🤗Transformers\
**Created:** [March 16, 2023, 12:44am UTC](https://discuss.huggingface.co/t/split-long-text-into-topics/33896 "2023-03-16T00:44:11Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![rioharper](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/rioharper/32/14751_2.png) [@rioharper](https://discuss.huggingface.co/u/rioharper)\
**Post date:** [March 16, 2023, 12:44am UTC](https://discuss.huggingface.co/t/split-long-text-into-topics/33896/1 "2023-03-16T00:44:11Z")

</div>

Hello!

So I am automating the process of creating datasets of interviews for the purposes of a chatbot. Often, topics of conversation (lets say of how their day was) spans multiple interactions, like so:

Person 1 (P1): So, how was your day?  
Person 2 (P2): Fine, but I had a long day of work. You?  
P1: Thats too bad, I had paid time off.  
P2: Thats lucky.  
P1: You watch that new game on Saturday?

I would like to be able to identify and split conversations up into these sort of segments about a focuses topics, so that the last interaction (P2) signifies a new topic. Obviously, in normal conversation, these types of things are a little more muddy than the sample I provided so I want to avoid traditional “find a certain character/keyphrase” method to split topics up instead.

I just wanted to check if a model like (or one that I could relativity quickly modify) that fits those parameters exist. If not, I’m sure its not a huge task to build one myself. But if I don’t have to do more than I need to, it’s worth a shot asking here.

Thanks!