Interactive Classification Training
What It Does
This app helps you train a text classification model step by step. You upload a CSV file, choose the text column, enter the labels you want to use, and start from a Hugging Face model.
The app then shows one text at a time. It suggests a label, and you can accept a category, choose another one, go back, or defer the example. Once you have labelled enough examples, you can train a model, download the labelled CSV, and download the trained model.
When To Use It
Use this app when you want to build a small custom classifier for your own research categories.
Research Examples
- Train a classifier to recognise themes such as sentiment, topic, genre, argument type, or stance in a corpus.
- Build a project-specific classifier for manually defined categories, such as types of policy claims, interview responses, or discourse strategies.
Good To Know
The quality of the classifier depends on your labels, the amount of labelled data, and the suitability of the starting model. The app is useful for iterative experimentation, but important research results should still be validated carefully.
You can find this app in the B-TXT toolsuite under Interactive Apps on the VUB-HPC portal.