Word Prediction Shiny App

Marta Z. Naves Sánchez
05/05/2026

Overview of the Application

Slide 1: Project Overview

This project is a Next Word Prediction Shiny Application.

Built using R and Shiny
Uses an n-gram language model
Predicts the most likely next word based on user input
Designed as an interactive text prediction tool similar to mobile keyboards

Slide 2: Algorithm Description

The prediction system is based on a backoff n-gram model:

4-gram → 3-gram → 2-gram → unigram fallback
Uses frequency counts from a large text corpus (Twitter, blogs, news)
Applies weighted scoring:
- Higher weight to longer context (4-gram strongest)
- Smoothing using log-frequency
- Unigram used only when no context match exists

This ensures:

Context awareness
Stability in prediction
Reduced randomness

Slide 3: App Functionality

The Shiny app includes:

Text input box for user phrase
Slider to select number of predictions
“Predict” button for execution
Output of top predicted words
Test dataset for evaluation of performance

How it works:

1. User enters a phrase
2. System extracts last words
3. Matches n-grams
4. Returns ranked predictions

Slide 4: User Experience

The app behaves like a lightweight predictive keyboard.

Experience characteristics:

Fast response time (~instant prediction)
Simple and clean interface
Context-aware suggestions
Sometimes general words appear when context is limited

Strength:

Easy to use and intuitive

Limitation:

Performance depends on training data quality and size

Slide 5: Evaluation & Conclusion

Strengths:

Clear probabilistic language model
Interactive Shiny interface
Real-time prediction capability
Lightweight and deployable on shinyapps.io

Novelty:

Combines classical NLP (n-grams) with interactive UI
Demonstrates full pipeline: data → model → app

This project demonstrates:

Solid understanding of NLP fundamentals
Ability to build and deploy data products
Practical engineering with R Shiny