Problem & Goal

• Predict the next word given a short phrase
• Useful for smart keyboards and text completion
• Goal: build a fast and simple prediction product

Data

• English text from:
  • News articles
  • Blogs
  • Twitter posts
• Preprocessing:
  • Lowercasing
  • Removing punctuation and URLs
  • Tokenization

Prediction Algorithm

• N-gram language model (2-gram, 3-gram, 4-gram)
• Backoff strategy:
  • 4-gram → 3-gram → 2-gram → unigram
• Frequency-based lookup for fast prediction

Shiny Application

• Text input for entering a phrase
• Submit button to trigger prediction
• Outputs a single predicted word
• Designed for fast response time

User Experience & Value

• Simple and intuitive interface
• Always returns a prediction
• Lightweight and fast
• Future improvements:
  • Multiple predictions
  • Profanity filtering
  • Smoothing techniques