Update README.md
Browse files
README.md
CHANGED
@@ -75,7 +75,6 @@ Que los estudiantes vuelven a clase.
|
|
75 |
|
76 |
|
77 |
# Data explanation
|
78 |
-
|
79 |
- **web_url** (int): The URL of the news article
|
80 |
- **web_headline** (str): The headline of the article, which is a Clickbait.
|
81 |
- **web_text** (int): The body of the article.
|
@@ -83,11 +82,107 @@ Que los estudiantes vuelven a clase.
|
|
83 |
- **summary** (str): The summary written by humans that answers the clickbait headline.
|
84 |
|
85 |
# Dataset Description
|
86 |
-
|
87 |
- **Curated by:** [Iker García-Ferrero](https://ikergarcia1996.github.io/Iker-Garcia-Ferrero/)
|
88 |
- **Language(s) (NLP):** Spanish
|
89 |
- **License:** apache-2.0
|
90 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
91 |
# Uses
|
92 |
This dataset is intended to build models tailored for academic research that can extract information from large texts. The objective is to research whether current LLMs, given a question formulated as a Clickbait headline, can locate the answer within the article body and summarize the information in a few words. The dataset also aims to serve as a task to evaluate the performance of current LLMs in Spanish.
|
93 |
|
@@ -95,13 +190,11 @@ This dataset is intended to build models tailored for academic research that can
|
|
95 |
You cannot use this dataset to develop systems that directly harm the newspapers included in the dataset. This includes using the dataset to train profit-oriented LLMs capable of generating articles from a short text or headline, as well as developing profit-oriented bots that automatically summarize articles without the permission of the article's owner. Additionally, you are not permitted to train a system with this dataset that generates clickbait headlines.
|
96 |
|
97 |
# Dataset Creation
|
98 |
-
|
99 |
The dataset has been meticulously created by hand. We utilize two sources to compile Clickbait articles:
|
100 |
- The Twitter user [@ahorrandoclick1](https://twitter.com/ahorrandoclick1), who reposts Clickbait articles along with a hand-crafted summary. Although we use their summaries as a reference, most of them have been rewritten (750 examples from this source).
|
101 |
- The web demo [⚔️ClickbaitFighter⚔️](https://iker-clickbaitfighter.hf.space/), which operates a pre-trained model using an early iteration of our dataset. We collect all the model inputs/outputs and manually correct them (100 examples from this source).
|
102 |
|
103 |
# Who are the annotators?
|
104 |
-
|
105 |
The dataset was annotated by [Iker García-Ferrero](https://ikergarcia1996.github.io/Iker-Garcia-Ferrero/) and validated by .
|
106 |
The annotation took ~40 hours.
|
107 |
|
|
|
75 |
|
76 |
|
77 |
# Data explanation
|
|
|
78 |
- **web_url** (int): The URL of the news article
|
79 |
- **web_headline** (str): The headline of the article, which is a Clickbait.
|
80 |
- **web_text** (int): The body of the article.
|
|
|
82 |
- **summary** (str): The summary written by humans that answers the clickbait headline.
|
83 |
|
84 |
# Dataset Description
|
|
|
85 |
- **Curated by:** [Iker García-Ferrero](https://ikergarcia1996.github.io/Iker-Garcia-Ferrero/)
|
86 |
- **Language(s) (NLP):** Spanish
|
87 |
- **License:** apache-2.0
|
88 |
|
89 |
+
# Dataset Usage
|
90 |
+
|
91 |
+
1. The easiest way to evaluate an LLM with this dataset if using the Language Model Evaluation Harness library: https://github.com/EleutherAI/lm-evaluation-harness
|
92 |
+
|
93 |
+
```bash
|
94 |
+
```
|
95 |
+
|
96 |
+
2. If you want to train an LLM or reproduce the results in our paper, you can use our code. See the repository for more info: [https://github.com/ikergarcia1996/NoticIA](https://github.com/ikergarcia1996/NoticIA)
|
97 |
+
|
98 |
+
3. If you want to manually load the dataset and run inference with an LLM:
|
99 |
+
You can load the dataset with the following command:
|
100 |
+
```Python
|
101 |
+
from datasets import load_dataset
|
102 |
+
dataset = load_dataset("Iker/NoticIA")
|
103 |
+
```
|
104 |
+
|
105 |
+
In order to perform inference with LLMs, you need to build a prompt. The one we use in our paper is:
|
106 |
+
```Python
|
107 |
+
def clickbait_prompt(
|
108 |
+
headline: str,
|
109 |
+
body: str,
|
110 |
+
) -> str:
|
111 |
+
"""
|
112 |
+
Generate the prompt for the model.
|
113 |
+
Args:
|
114 |
+
headline (`str`):
|
115 |
+
The headline of the article.
|
116 |
+
body (`str`):
|
117 |
+
The body of the article.
|
118 |
+
Returns:
|
119 |
+
`str`: The formatted prompt.
|
120 |
+
"""
|
121 |
+
return (
|
122 |
+
f"Ahora eres una Inteligencia Artificial experta en desmontar titulares sensacionalistas o clickbait. "
|
123 |
+
f"Tu tarea consiste en analizar noticias con titulares sensacionalistas y "
|
124 |
+
f"generar un resumen de una sola frase que revele la verdad detrás del titular.\n"
|
125 |
+
f"Este es el titular de la noticia: {headline}\n"
|
126 |
+
f"El titular plantea una pregunta o proporciona información incompleta. "
|
127 |
+
f"Debes buscar en el cuerpo de la noticia una frase que responda lo que se sugiere en el título. "
|
128 |
+
f"Responde siempre que puedas parafraseando el texto original. "
|
129 |
+
f"Usa siempre las mínimas palabras posibles. "
|
130 |
+
f"Recuerda responder siempre en Español.\n"
|
131 |
+
f"Este es el cuerpo de la noticia:\n"
|
132 |
+
f"{body}\n"
|
133 |
+
)
|
134 |
+
```
|
135 |
+
|
136 |
+
Here is a practical end-to-end example using the text generation pipeline.
|
137 |
+
```python
|
138 |
+
from transformers import pipeline
|
139 |
+
from datasets import load_dataset
|
140 |
+
|
141 |
+
generator = pipeline(model="google/gemma-2b-it",device_map="auto")
|
142 |
+
dataset = load_dataset("Iker/NoticIA")
|
143 |
+
|
144 |
+
example = dataset["test"][0]
|
145 |
+
prompt = clickbait_prompt(headline=example["web_headline"],body=example["web_text"])
|
146 |
+
outputs = generator(prompt, return_full_text=False,max_length=4096)
|
147 |
+
print(outputs)
|
148 |
+
|
149 |
+
# [{'generated_text': 'La tuitera ha recibido un número considerable de comentarios y mensajes de apoyo.'}]
|
150 |
+
```
|
151 |
+
|
152 |
+
Here is a practical end-to-end example using the generate function
|
153 |
+
```python
|
154 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
155 |
+
from datasets import load_dataset
|
156 |
+
|
157 |
+
tokenizer = AutoTokenizer.from_pretrained("google/gemma-2b-it")
|
158 |
+
model = AutoModelForCausalLM.from_pretrained("google/gemma-2b-it",device_map="auto",quantization_config={"load_in_4bit": True})
|
159 |
+
dataset = load_dataset("Iker/NoticIA")
|
160 |
+
|
161 |
+
example = dataset["test"][0]
|
162 |
+
prompt = clickbait_prompt(headline=example["web_headline"],body=example["web_text"])
|
163 |
+
prompt = tokenizer.apply_chat_template(
|
164 |
+
[{"role": "user", "content": prompt}],
|
165 |
+
tokenize=False,
|
166 |
+
add_generation_prompt=True,
|
167 |
+
)
|
168 |
+
model_inputs = tokenizer(
|
169 |
+
text=prompt,
|
170 |
+
max_length=3096,
|
171 |
+
truncation=True,
|
172 |
+
padding=False,
|
173 |
+
return_tensors="pt",
|
174 |
+
add_special_tokens=False,
|
175 |
+
)
|
176 |
+
|
177 |
+
outputs = model.generate(**model_inputs,max_length=4096)
|
178 |
+
output_text = tokenizer.batch_decode(outputs)
|
179 |
+
|
180 |
+
print(output_text[0])
|
181 |
+
|
182 |
+
# La usuaria ha comprado un abrigo para su abuela de 97 años, pero la "yaya" no está de acuerdo.
|
183 |
+
```
|
184 |
+
|
185 |
+
|
186 |
# Uses
|
187 |
This dataset is intended to build models tailored for academic research that can extract information from large texts. The objective is to research whether current LLMs, given a question formulated as a Clickbait headline, can locate the answer within the article body and summarize the information in a few words. The dataset also aims to serve as a task to evaluate the performance of current LLMs in Spanish.
|
188 |
|
|
|
190 |
You cannot use this dataset to develop systems that directly harm the newspapers included in the dataset. This includes using the dataset to train profit-oriented LLMs capable of generating articles from a short text or headline, as well as developing profit-oriented bots that automatically summarize articles without the permission of the article's owner. Additionally, you are not permitted to train a system with this dataset that generates clickbait headlines.
|
191 |
|
192 |
# Dataset Creation
|
|
|
193 |
The dataset has been meticulously created by hand. We utilize two sources to compile Clickbait articles:
|
194 |
- The Twitter user [@ahorrandoclick1](https://twitter.com/ahorrandoclick1), who reposts Clickbait articles along with a hand-crafted summary. Although we use their summaries as a reference, most of them have been rewritten (750 examples from this source).
|
195 |
- The web demo [⚔️ClickbaitFighter⚔️](https://iker-clickbaitfighter.hf.space/), which operates a pre-trained model using an early iteration of our dataset. We collect all the model inputs/outputs and manually correct them (100 examples from this source).
|
196 |
|
197 |
# Who are the annotators?
|
|
|
198 |
The dataset was annotated by [Iker García-Ferrero](https://ikergarcia1996.github.io/Iker-Garcia-Ferrero/) and validated by .
|
199 |
The annotation took ~40 hours.
|
200 |
|