any-news-classifier

This model is a fine-tuned version of sentence-transformers/LaBSE on my news dataset. The learning news dataset is a well-balanced sample of recent news from the last five years.

It achieves the following results on the evaluation set:

• Loss: 0.3820
• Accuracy: 0.9029
• F1: 0.9025
• Precision: 0.9030
• Recall: 0.9029

Model description

This is a multi-class classifier of Russian news, made with the LaBSE model finetune for AntiSMI Project. The news category is assigned by the classifier to one of 11 categories:

• climate (климат)
• conflicts (конфликты)
• culture (культура)
• economy (экономика)
• gloss (глянец)
• health (здоровье)
• politics (политика)
• science (наука)
• society (общество)
• sports (спорт)
• travel (путешествия)

Testing this model on Spaces

You can try the model and evaluate its quality here

How to use

from transformers import pipeline

category_mapper = {
'LABEL_0': 'climate',
'LABEL_1': 'conflicts',
'LABEL_2': 'culture',
'LABEL_3': 'economy',
'LABEL_4': 'gloss',
'LABEL_5': 'health',
'LABEL_6': 'politics',
'LABEL_7': 'science',
'LABEL_8': 'society',
'LABEL_9': 'sports',
'LABEL_10': 'travel'
}

# Используйте предобученную модель из Hugging Face Hub
classifier = pipeline("text-classification", model="data-silence/rus-news-classifier")

def predict_category(text):
    result = classifier(text)
    category = category_mapper[result[0]['label']]
    score = result[0]['score']
    return category, score

predict_category("В Париже завершилась церемония закрытия Олимпийских игр")
# ('sports', 0.9959506988525391)

Intended uses & limitations

The "gloss" category is used to select yellow press, trashy and dubious news. The model can get confused in the classification of news categories politics, society and conflicts.

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
0.3544	1.0	3596	0.3517	0.8861	0.8860	0.8915	0.8861
0.2738	2.0	7192	0.3190	0.8995	0.8987	0.9025	0.8995
0.19	3.0	10788	0.3524	0.9016	0.9015	0.9019	0.9016
0.1402	4.0	14384	0.3820	0.9029	0.9025	0.9030	0.9029
0.1055	5.0	17980	0.4399	0.9022	0.9018	0.9024	0.9022

Framework versions

• Transformers 4.42.4
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1