# Unlabelled zero-shot-classification

**URL:** </static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Funlabelled-zero-shot-classification%2F41028%3E%5C
**Category:** 🤗Transformers\
**Created:** [May 26, 2023, 1:23am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Funlabelled-zero-shot-classification%2F41028 "2023-05-26T01:23:16Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![francisurias](https://avatars.discourse-cdn.com/v4/letter/f/b3f665/32.png) [@francisurias](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Ffrancisurias)%5C
**Post date:** [May 26, 2023, 1:23am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Funlabelled-zero-shot-classification%2F41028%2F1 "2023-05-26T01:23:16Z")

</div>

Hi,

I am working on a project that aims to classify surveys into categories, and I’m currently using zero-shot-classification for this. Using zero-shot-classification, I am able to specify the labels I want it to classify into, but is there a way to make it so that I do not need to specify the labels, and the model auto labels the content? What would I need to do to achieve this? (Train the model? Models specifically for this?)

---

<div class="post-metadata">

**Author:** ![mirix](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/mirix/32/16327_2.png) [@mirix](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fmirix)%5C
**Post date:** [May 26, 2023, 7:12am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Funlabelled-zero-shot-classification%2F41028%2F2 "2023-05-26T07:12:54Z")

</div>

Perhaps automatic extraction of the labels first?

> **[GitHub - MaartenGr/BERTopic: Leveraging BERT and c-TF-IDF to create easily...](https://github.com/MaartenGr/BERTopic)**
>
> Leveraging BERT and c-TF-IDF to create easily interpretable topics. - GitHub - MaartenGr/BERTopic: Leveraging BERT and c-TF-IDF to create easily interpretable topics.

I must confess that, even after a lot of parametrisation, this did not work well for me on a very messy email dataset.

What I ended up doing was fine-tuning a pre-trained model for keyword extraction. Not great either, but somehow better.

You can then couple that with your approach…

I am an absolute beginner in this field, by the way.