The Problem

• Typing on phones is slow and error-prone.
• Predictive text saves time and reduces typos.
• Goal: predict the next word given a phrase.

The Algorithm

• Trained on blogs, news, and Twitter (~150,000 lines sampled).
• Text cleaned: lowercase, punctuation removed, non-ASCII stripped.
• Built a trigram back-off model:
  • Try trigram match first
  • Fall back to bigram
  • Fall back to most common unigram

The App

• Simple text box: type a phrase.
• Click Predict to see the next word.
• Fast (< 1 second) because the model is precomputed and small.

How to Use It

1. Open the app: https://alexbrijo.shinyapps.io/wordpredictor/
2. Type a phrase like “I love”
3. Click Predict
4. The app returns the most likely next word.

• Try: “happy birthday to” → “you”
• Try: “good morning” → “to”

Why It Works

• Lightweight model → fast and free to host.
• Back-off strategy handles unseen phrases gracefully.
• Trained on real-world text: tweets, blogs, news.
• Easily extendable to 4-grams or 5-grams.

Thank You