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!