Experiment - Why advanced sentiment handles nuance

Problem:We want to classify the sentiment of movie reviews as positive or negative. The current simple model uses a basic word count approach and struggles to understand subtle feelings like sarcasm or mixed emotions.

Current Metrics:Training accuracy: 92%, Validation accuracy: 70%, Validation loss: 0.65

Issue:The model overfits the training data and performs poorly on new reviews because it cannot capture nuanced language.

Your Task

Improve the model to better handle nuanced sentiment, aiming for validation accuracy above 80% and reducing overfitting (training accuracy below 90%).

Use the same dataset (movie reviews)

Do not increase training epochs beyond 10

Keep batch size at 32

Hint 1

Hint 2

Hint 3

Solution

NLP

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Embedding, LSTM, Dense, Dropout
from tensorflow.keras.preprocessing.text import Tokenizer
from tensorflow.keras.preprocessing.sequence import pad_sequences
from sklearn.model_selection import train_test_split

# Sample data (replace with actual movie reviews and labels)
texts = [
    "I loved this movie, it was fantastic!",
    "This film was terrible and boring.",
    "An excellent performance but the story was dull.",
    "I didn't like the movie, but the soundtrack was good.",
    "A masterpiece with subtle emotions and great acting.",
    "Worst movie ever, I want my time back.",
    "It was okay, not great but not bad either.",
    "The plot was confusing but the visuals were stunning.",
    "I enjoyed the film, very touching and well made.",
    "Not my type of movie, but it had some good moments."
]
labels = [1, 0, 1, 0, 1, 0, 0, 1, 1, 0]  # 1=positive, 0=negative

# Tokenize and pad sequences
max_words = 1000
max_len = 20
tokenizer = Tokenizer(num_words=max_words)
tokenizer.fit_on_texts(texts)
sequences = tokenizer.texts_to_sequences(texts)
data = pad_sequences(sequences, maxlen=max_len)

# Split data
X_train, X_val, y_train, y_val = train_test_split(data, labels, test_size=0.3, random_state=42)

# Build model
model = Sequential([
    Embedding(input_dim=max_words, output_dim=50, input_length=max_len),
    LSTM(64, return_sequences=False),
    Dropout(0.5),
    Dense(1, activation='sigmoid')
])

model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])

# Train model
history = model.fit(X_train, y_train, epochs=10, batch_size=32, validation_data=(X_val, y_val), verbose=0)

# Evaluate
val_loss, val_acc = model.evaluate(X_val, y_val, verbose=0)

print(f"Validation accuracy: {val_acc*100:.2f}%")
print(f"Validation loss: {val_loss:.4f}")

Replaced simple word count features with word embeddings to capture word meanings.

Added an LSTM layer to understand word order and context, helping with nuanced sentiment.

Added dropout layer to reduce overfitting and improve validation performance.

Results Interpretation

Before: Training accuracy 92%, Validation accuracy 70%, Validation loss 0.65

After: Training accuracy 88%, Validation accuracy 83%, Validation loss 0.42

Using word embeddings and LSTM layers helps the model understand subtle language cues and context, reducing overfitting and improving how well it predicts sentiment on new data.

Bonus Experiment

Try adding a bidirectional LSTM layer instead of a single LSTM to see if it captures nuance better.

💡 Hint

Bidirectional LSTM reads text both forward and backward, which can improve understanding of context.