lstm-news-classifier

Bidirectional LSTM (Keras / TensorFlow) trained for news topic classification on the HuffPost News Category Dataset.

Model Details

Item	Value
Base architecture	Bidirectional LSTM (2 layers)
Task	Multi-class text classification
# Classes	12
Vocab size	60,000
Max token length	256
Embedding dim	128
Test Accuracy	0.6909
Test F1 (macro)	0.6191

Labels

• 0: business_tech
• 1: crime_weird
• 2: entertainment
• 3: environment
• 4: food_travel
• 5: health
• 6: home_style
• 7: lifestyle
• 8: news_politics
• 9: positive
• 10: science_edu
• 11: sports

Usage

import pickle, json, numpy as np
import tensorflow as tf
from tensorflow.keras.preprocessing.sequence import pad_sequences

lstm_model = tf.keras.models.load_model('lstm_model.keras')
with open('keras_tokenizer.pkl', 'rb') as f:
    tokenizer = pickle.load(f)
with open('label_meta.json') as f:
    meta = json.load(f)

ID2LABEL   = {int(k): v for k, v in meta['id2label'].items()}
MAX_LENGTH = meta['max_length']

def predict(text):
    seq    = tokenizer.texts_to_sequences([text])
    padded = pad_sequences(seq, maxlen=MAX_LENGTH, padding='post', truncating='post')
    probs  = lstm_model.predict(padded, verbose=0).squeeze()
    idx    = int(np.argmax(probs))
    return {'label': ID2LABEL[idx], 'score': float(probs[idx])}

print(predict('AI is transforming healthcare with new diagnostic tools'))

Training

• Optimizer: Adam, lr=1e-3 with ReduceLROnPlateau
• Epochs: 15 (best checkpoint by val_loss, EarlyStopping patience=4)
• Loss: sparse_categorical_crossentropy with class-weight balancing
• GPU: NVIDIA T4 (Kaggle)