News Classifier — new-clasification-manish

Yeh model news headlines aur articles ko automatically 6 categories mein classify karta hai.

Developed by: Manish Negi — Isuremedia
Base Model: distilbert-base-multilingual-cased
Languages: English + Hindi
Accuracy: 78% (Test Set)


Categories

ID Category Examples
0 politics Elections, Government, Parliament, Modi, BJP
1 sports Cricket, IPL, Football, Olympics
2 business Sensex, RBI, Stock Market, Economy
3 technology iPhone, AI, Gadgets, Software
4 entertainment Bollywood, Movies, Celebrities
5 international World News, USA, China, War

How to Use

Install

pip install transformers torch

Basic Usage

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="dev-isure/new-clasification-manish"
)

# English news
result = classifier("Virat Kohli scores century in IPL match")
print(result)
# [{'label': 'sports', 'score': 0.95}]

# Hindi news
result = classifier("सेंसेक्स 500 अंक गिरा, बाजार में भारी बिकवाली")
print(result)
# [{'label': 'business', 'score': 0.91}]

Multiple Articles

news = [
    "PM Modi inaugurates new expressway in UP",
    "India beats Australia in T20 series",
    "RBI keeps repo rate unchanged at 6.5%",
    "Apple launches new iPhone 17 Pro",
    "Shah Rukh Khan's new film breaks box office records",
    "Russia Ukraine war ceasefire talks begin"
]

results = classifier(news)
for text, res in zip(news, results):
    print(f"{res['label']:<15} ({res['score']:.2f})  {text[:50]}")

Training Details

Detail Value
Base Model distilbert-base-multilingual-cased
Training Articles 792
Validation Articles 99
Test Articles 100
Epochs 3
Batch Size 8
Learning Rate 2e-5
Test Accuracy 78%

Training Data Sources

  • NDTV, Times of India, The Hindu, Hindustan Times
  • Economic Times, ESPN Cricinfo
  • BBC Hindi, Aaj Tak, Dainik Bhaskar
  • Gadgets360, Bollywood Hungama, Pinkvilla

Limitations

  • Health aur Crime categories is model mein nahi hain
  • Hinglish (Hindi-English mixed) pe accuracy thodi kam ho sakti hai
  • Zyada data se accuracy improve hogi
Downloads last month
7
Safetensors
Model size
0.1B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for dev-isure/new-clasification-manish

Finetuned
(442)
this model

Evaluation results