Interpress Turkish News Classification

🚀 INTERPRESS NEWS CLASSIFICATION

This project focuses on classifying Interpress news using machine learning techniques, achieving high accuracy on real - world data.

🚀 Quick Start

The following sections will guide you through the dataset, model performance, and how to use the model with Torch and TensorFlow.

✨ Features

• Real - world data: Utilizes a real - world news dataset from Interpress.
• High accuracy: The model achieves 97% accuracy on both training and validation data.
• Multi - framework support: Can be used with both Torch and TensorFlow.

📦 Installation

To use this model, you need to install the transformers library. You can use the following command:

pip install transformers or pip install transformers==4.3.3

📚 Documentation

Dataset

The dataset is downloaded from Interpress. It is real - world data. Initially, there were 273K data, but I filtered them and used 108K data for this model. For more information about the dataset, please visit this link.

Model

The model's accuracy on the training data and validation data is 97%. The data is split into 80% for training and 20% for validation. The results are shown as follows:

Classification report

Confusion matrix

💻 Usage Examples

Usage for Torch

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("serdarakyol/interpress-turkish-news-classification")
model = AutoModelForSequenceClassification.from_pretrained("serdarakyol/interpress-turkish-news-classification")

import torch

if torch.cuda.is_available():    
    device = torch.device("cuda")
    model = model.cuda()
    print('There are %d GPU(s) available.' % torch.cuda.device_count())
    print('GPU name is:', torch.cuda.get_device_name(0))
else:
    print('No GPU available, using the CPU instead.')
    device = torch.device("cpu")

import numpy as np

def prediction(news):
    news=[news]
    indices=tokenizer.batch_encode_plus(
    news,
    max_length=512,
    add_special_tokens=True,
    return_attention_mask=True,
    padding='max_length',
    truncation=True,
    return_tensors='pt')

    inputs = indices["input_ids"].clone().detach().to(device)
    masks = indices["attention_mask"].clone().detach().to(device)

    with torch.no_grad():
        output = model(inputs, token_type_ids=None,attention_mask=masks)

    logits = output[0]
    logits = logits.detach().cpu().numpy()
    pred = np.argmax(logits,axis=1)[0]
    return pred

news = r"ABD'den Prens Selman'a yaptırım yok Beyaz Saray Sözcüsü Psaki, Muhammed bin Selman'a yaptırım uygulamamanın \"doğru karar\" olduğunu savundu. Psaki, \"Tarihimizde, Demokrat ve Cumhuriyetçi başkanların yönetimlerinde diplomatik ilişki içinde olduğumuz ülkelerin liderlerine yönelik yaptırım getirilmemiştir\" dedi."

You can find the news in this link (news date: 02/03/2021)

labels = {
    0 : "Culture-Art",
    1 : "Economy",
    2 : "Politics",
    3 : "Education",
    4 : "World",
    5 : "Sport",
    6 : "Technology",
    7 : "Magazine",
    8 : "Health",
    9 : "Agenda"
}
pred = prediction(news)
print(labels[pred])
# > World

Usage for TensorFlow

import tensorflow as tf
from transformers import BertTokenizer, TFBertForSequenceClassification
import numpy as np

tokenizer = BertTokenizer.from_pretrained('serdarakyol/interpress-turkish-news-classification')
model = TFBertForSequenceClassification.from_pretrained("serdarakyol/interpress-turkish-news-classification")

inputs = tokenizer(news, return_tensors="tf")
inputs["labels"] = tf.reshape(tf.constant(1), (-1, 1)) # Batch size 1

outputs = model(inputs)
loss = outputs.loss
logits = outputs.logits
pred = np.argmax(logits,axis=1)[0]
labels[pred]
# > World

Acknowledgments

Thanks to @yavuzkomecoglu for contributions.

If you have any questions, please don't hesitate to contact me.