|
|
--- |
|
|
library_name: transformers |
|
|
tags: |
|
|
- clickbait |
|
|
- headline-generation |
|
|
- summarization |
|
|
- seq2seq |
|
|
license: mit |
|
|
language: en |
|
|
--- |
|
|
|
|
|
# Model Card for `bwlw3127/clickbait-bart-dailymail` |
|
|
|
|
|
This model fine-tunes [`facebook/bart-base`](https://huggingface.co/facebook/bart-base) to generate **clickbait-style headlines** from DailyMail articles. |
|
|
It takes an article as input and produces a short, attention-grabbing headline. |
|
|
|
|
|
--- |
|
|
|
|
|
## Model Details |
|
|
|
|
|
### Model Description |
|
|
|
|
|
This is a sequence-to-sequence model trained with the 🤗 Transformers library. It was adapted from `facebook/bart-base` and fine-tuned on scraped DailyMail article/headline pairs. |
|
|
The goal is to demonstrate how large language models can be steered toward stylistic objectives such as “clickbait” headline generation. |
|
|
|
|
|
- **Developed by:** @bwlw3127 |
|
|
- **Model type:** Seq2Seq (Encoder–Decoder) — BART |
|
|
- **Language(s):** English |
|
|
- **License:** MIT (code); base model follows [facebook/bart-base license](https://huggingface.co/facebook/bart-base) |
|
|
- **Finetuned from:** `facebook/bart-base` |
|
|
|
|
|
### Model Sources |
|
|
|
|
|
- **Repository (code):** [GitHub – weiw3127/clickbait](https://github.com/weiw3127/clickbait) |
|
|
- **Model (weights):** [Hugging Face Hub](https://huggingface.co/bwlw3127/clickbait-bart-dailymail) |
|
|
|
|
|
--- |
|
|
|
|
|
## Uses |
|
|
|
|
|
### Direct Use |
|
|
|
|
|
- Generate sensational or catchy headlines from article text. |
|
|
- Educational demo for seq2seq fine-tuning with Hugging Face. |
|
|
|
|
|
### Downstream Use |
|
|
|
|
|
- Adaptation to other headline-generation tasks (news summarization, marketing, etc.) by re-finetuning on different headline corpora. |
|
|
|
|
|
### Out-of-Scope Use |
|
|
|
|
|
- Producing factually reliable news summaries. |
|
|
- Sensitive domains where misleading/sensational output may cause harm. |
|
|
|
|
|
--- |
|
|
|
|
|
## Bias, Risks, and Limitations |
|
|
|
|
|
- The model reflects biases in DailyMail content and clickbait style. |
|
|
- Headlines may exaggerate, distort, or misrepresent facts. |
|
|
- It should not be used in production systems where factual accuracy is critical. |
|
|
|
|
|
### Recommendations |
|
|
|
|
|
- Use only in controlled or educational contexts. |
|
|
- Always review outputs manually before publishing. |
|
|
|
|
|
--- |
|
|
|
|
|
## How to Get Started with the Model |
|
|
|
|
|
```python |
|
|
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM |
|
|
|
|
|
model_id = "bwlw3127/clickbait-bart-dailymail" |
|
|
tokenizer = AutoTokenizer.from_pretrained(model_id) |
|
|
model = AutoModelForSeq2SeqLM.from_pretrained(model_id) |
|
|
|
|
|
article = "Scientists have discovered a new exoplanet twice the size of Earth..." |
|
|
inputs = tokenizer("generate clickbait headline: " + article, |
|
|
return_tensors="pt", truncation=True, max_length=512) |
|
|
|
|
|
outputs = model.generate(**inputs, max_length=24, num_beams=4, early_stopping=True) |
|
|
print(tokenizer.decode(outputs[0], skip_special_tokens=True)) |
|
|
|
