Spanish News Classification Headlines

SNCH: this model was developed by M47Labs the goal is text classification, the base model use was BETO, however this model has not been fine-tuned on any dataset. The objective is to show the performance of this model when is used with the objective of inference without training at all.

Dataset validation Sample

Dataset size : 1000

Columns: idTask,task content 1,idTag,tag.

Labels:

• ciencia_tecnologia

• clickbait

• cultura

• deportes

• economia

• educacion

• medio_ambiente

• opinion

• politica

• sociedad

Example of Use

Pipeline

import torch
from transformers import AutoTokenizer, BertForSequenceClassification,TextClassificationPipeline


review_text = 'los vehiculos que esten esperando pasajaeros deberan estar apagados para reducir emisiones'
path = "M47Labs/spanish_news_classification_headlines_untrained"
tokenizer = AutoTokenizer.from_pretrained(path)
model = BertForSequenceClassification.from_pretrained(path)


nlp = TextClassificationPipeline(task = "text-classification",
                model = model,
                tokenizer = tokenizer)

print(nlp(review_text))

[{'label': 'medio_ambiente', 'score': 0.2834321384291023}]

Pytorch

import torch
from transformers import AutoTokenizer, BertForSequenceClassification,TextClassificationPipeline
from numpy import np

model_name  = 'M47Labs/spanish_news_classification_headlines_untrained'
MAX_LEN = 32


tokenizer = AutoTokenizer.from_pretrained(model_name)

model = AutoModelForSequenceClassification.from_pretrained(model_name)

texto = "las emisiones estan bajando, debido a las medidas ambientales tomadas por el gobierno"


encoded_review = tokenizer.encode_plus(
  texto,
  max_length=MAX_LEN,
  add_special_tokens=True,
  #return_token_type_ids=False,
  pad_to_max_length=True,
  return_attention_mask=True,
  return_tensors='pt',
)

input_ids = encoded_review['input_ids']
attention_mask = encoded_review['attention_mask']
output = model(input_ids, attention_mask)

_, prediction = torch.max(output['logits'], dim=1)
print(f'Review text: {texto}')

print(f'Sentiment  : {model.config.id2label[prediction.detach().cpu().numpy()[0]]}')

Review text: las emisiones estan bajando, debido a las medidas ambientales tomadas por el gobierno

Sentiment : opinion

A more in depth example on how to use the model can be found in this colab notebook: https://colab.research.google.com/drive/1XsKea6oMyEckye2FePW_XN7Rf8v41Cw_?usp=sharing

Validation Results