Non Factoid Question Category classification in English

NFQA model

Repository: https://github.com/Lurunchik/NF-CATS
Model trained with NFQA dataset. Base model is roberta-base-squad2, a RoBERTa-based model for the task of Question Answering, fine-tuned using the SQuAD2.0 dataset.
Uses NOT-A-QUESTION, FACTOID, DEBATE, EVIDENCE-BASED, INSTRUCTION, REASON, EXPERIENCE, COMPARISON labels.

How to use NFQA cat with HuggingFace

Load NFQA cat and its tokenizer:

from transformers import AutoTokenizer<br /> from nfqa_model import RobertaNFQAClassification<br /> nfqa_model = RobertaNFQAClassification.from_pretrained("Lurunchik/nf-cats")<br /> nfqa_tokenizer = AutoTokenizer.from_pretrained("deepset/roberta-base-squad2")<br />

Make prediction using helper function:

def get_nfqa_category_prediction(text):<br /> output = nfqa_model(**nfqa_tokenizer(text, return_tensors="pt"))<br /> index = output.logits.argmax()<br /> return nfqa_model.config.id2label[int(index)]<br /> get_nfqa_category_prediction('how to assign category?')<br /> # result<br /> #'INSTRUCTION'<br />

Demo

You can test the model via hugginface space.

Citation

If you use NFQA-cats in your work, please cite this paper
@misc{bolotova2022nfcats,<br /> author = {Bolotova, Valeriia and Blinov, Vladislav and Scholer, Falk and Croft, W. Bruce and Sanderson, Mark},<br /> title = {A Non-Factoid Question-Answering Taxonomy},<br /> year = {2022},<br /> isbn = {9781450387323},<br /> publisher = {Association for Computing Machinery},<br /> address = {New York, NY, USA},<br /> url = {https://doi.org/10.1145/3477495.3531926},<br /> doi = {10.1145/3477495.3531926},<br /> booktitle = {Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval},<br /> pages = {1196–1207},<br /> numpages = {12},<br /> keywords = {question taxonomy, non-factoid question-answering, editorial study, dataset analysis},<br /> location = {Madrid, Spain},<br /> series = {SIGIR '22}<br /> }<br />

Enjoy! 🤗