bert.py

import pandas as pd
from sklearn.model_selection import train_test_split
from transformers import BertTokenizer, TFBertForSequenceClassification
import tensorflow as tf

# Load preprocessed data
def load_data(file_path="preprocessed_reviews.csv"):
    return pd.read_csv("preprocessed_reviews.csv")

# Tokenize text using BERT tokenizer

def tokenize_text(tokenizer, texts, max_length):
    encodings = tokenizer(texts.tolist(), padding=True, truncation=True, max_length=max_length, return_tensors="tf")
    # Convert BatchEncoding to dictionary
    encodings_dict = {key: value.numpy() for key, value in encodings.items()}
    return encodings_dict


if __name__ == "__main__":
    # Load preprocessed data
    data = load_data("preprocessed_reviews.csv")

    # Check if 'sentiment' column exists
    if 'sentiment' in data.columns:
        # Split data into train and validation sets
        train_data, val_data = train_test_split(data, test_size=0.2, random_state=42)

        # Tokenize text using BERT tokenizer
        max_length = 128
        tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
        train_inputs = tokenize_text(tokenizer, train_data['clean_text'], max_length)
        val_inputs = tokenize_text(tokenizer, val_data['clean_text'], max_length)

        # Convert 'sentiment' column to numerical format
        num_labels = len(data['sentiment'].unique())
        train_labels = train_data['sentiment'].astype('category').cat.codes.values
        val_labels = val_data['sentiment'].astype('category').cat.codes.values

        # Fine-tuning BERT model for sequence classification
        model = TFBertForSequenceClassification.from_pretrained('bert-base-uncased', num_labels=num_labels)
        optimizer = tf.keras.optimizers.Adam(learning_rate=2e-5)
        loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
        metrics = ['accuracy']
        model.compile(optimizer=optimizer, loss=loss, metrics=metrics)

        # Train the model
        history = model.fit(
            train_inputs,
            train_labels,
            validation_data=(val_inputs, val_labels),
            epochs=3,
            batch_size=32,
            verbose=1
        )

        # Evaluate the model
        loss, accuracy = model.evaluate(val_inputs, val_labels)
        print(f'Validation loss: {loss}, Validation accuracy: {accuracy}')

        # Save the trained model
        model.save_pretrained('fine_tuned_bert_model')

    else:
        raise ValueError("The 'sentiment' column is not found in the DataFrame.")