NextWord Predictor: Smart Text Prediction Engine
Arun
2026-09-15
Slide 1: Executive Summary & Problem
- The Problem: Mobile typing is slow and prone to errors.
- The Solution: A real-time natural language prediction app.
- The Value: Saves keystrokes and improves typing speed.
Slide 2: Data Preprocessing & N-Grams
- Cleaned the SwiftKey US corpus (blogs, news, Twitter).
- Tokenized unigrams, bigrams, trigrams, and quadgrams.
- Pruned low-frequency terms to reduce memory footprint.
Slide 3: Prediction Algorithm
- Implements a Stupid Backoff / Katz Backoff routine.
- Checks highest-order n-gram match first (trigram/quadgram).
- Backs off to bigram and unigram if no match exists.
Slide 4: Shiny App Features
- Clean text-input interface with instant response.
- Shows top predicted word dynamically.
- Handles unknown tokens and typos gracefully.
Slide 5: Performance & Future Work
- Fast response time (<50ms) suitable for mobile/web.
- Lightweight memory footprint compatible with shinyapps.io.
- Future plan: Context-aware deep learning embeddings.
Thank You!