💬 Natural Language Processing

🔧 NLP Pipeline (Simplified)

• Tokenization – split text into words or sub‑words.
• Embeddings – convert tokens to numerical vectors (e.g., Word2Vec, GloVe).
• Sequence model – process sequences (RNN, LSTM, Transformer).
• Output – classification (sentiment, topic) or generation.

🚀 Live Demo: Sentiment Analysis

Enter any text (e.g., movie review, product feedback). The model (LSTM trained on IMDB reviews) predicts positive or negative sentiment.

// Model is a pre-trained LSTM (embedding + LSTM + dense)
// It was trained on 50,000 IMDB movie reviews.

// Preprocessing: tokenize, pad sequences to length 100
const model = await tf.loadLayersModel(modelUrl);
const inputSequence = tokenize(text, vocab);
const padded = padSequences([inputSequence], 100);
const prediction = model.predict(padded);
// output: probability of positive (0..1)

🌍 NLP in the Wild

• 💬 Chatbots and virtual assistants (Siri, Alexa, ChatGPT)
• 📊 Sentiment analysis (brand monitoring, social media)
• 🌐 Machine translation (Google Translate)
• 📄 Text summarisation (news, research)
• 🔍 Information retrieval (search engines)

✨ Challenge: Test Different Texts

Try these examples and observe the output:

• "I absolutely loved this product! It exceeded all expectations." (positive)
• "Terrible service, very disappointing." (negative)
• "The food was okay, nothing special." (neutral / mixed)