|
--- |
|
license: mit |
|
language: |
|
- en |
|
pipeline_tag: text2text-generation |
|
--- |
|
# News2Topic-V2-Flan-T5-base |
|
|
|
## Model Details |
|
|
|
- **Model type:** Text-to-Text Generation |
|
- **Language(s) (NLP):** English |
|
- **License:** MIT License |
|
- **Finetuned from model:** FLAN-T5 Base Model (Google AI) |
|
|
|
## Uses |
|
|
|
The News2Topic Flan T5-base model is designed for automatic generation of topic names from news articles or news-like text. It can be integrated into news aggregation platforms, content management systems, or used for enhancing news browsing and searching experiences by providing concise topics. |
|
|
|
## How to Get Started with the Model |
|
|
|
``` |
|
from transformers import pipeline |
|
|
|
pipe = pipeline("text2text-generation", model="textgain/News2Topic-V2-Flan-T5-base") |
|
|
|
news_text = "Your news text here." |
|
print(pipe(news_text)) |
|
``` |
|
|
|
## Training Details |
|
|
|
The News2Topic V2 Flan T5-base model was trained on a 20K sample of the "Newsroom" dataset (https://lil.nlp.cornell.edu/newsroom/index.html), annotated with data generated by a fine-tuned GPT-3.5-turbo on synthetic curated data. |
|
|
|
The model was trained for 10 epochs, with a learning rate of 0.00001, a maximum sequence length of 512, and a training batch size of 12. |
|
|
|
## Citation |
|
|
|
**BibTeX:** |
|
``` |
|
@article{Kosar_DePauw_Daelemans_2024, |
|
title={Comparative Evaluation of Topic Detection: Humans vs. LLMs}, volume={13}, |
|
url={https://www.clinjournal.org/clinj/article/view/173}, journal={Computational Linguistics in the Netherlands Journal}, |
|
author={Kosar, Andriy and De Pauw, Guy and Daelemans, Walter}, |
|
year={2024}, |
|
month={Mar.}, |
|
pages={91–120} } |
|
``` |
