Introduction
This app predicts the next word based on a user-input phrase.
Built using R, Shiny, and NLP techniques.
Focused on real-time performance and ease of use.
Data & Preprocessing
- Dataset: Blogs, News, Twitter data from SwiftKey.
- Text cleaned: removed punctuation, numbers, stopwords.
- Created n-grams (1–4) using
tidytext and tokenizers.
Prediction Algorithm
- Used a Stupid Backoff model.
- App searches for matching 4-gram, then backs off to 3, 2, and 1.
- Efficient lookup with
data.table.
App Features
- Users enter a phrase like: “I love”
- App instantly predicts next likely word.
- Clean, fast, mobile-responsive UI using Shiny.
Business Use Case
- Ideal for mobile keyboards, email auto-complete, chatbot typing aids.
- Saves time and enhances user experience.
- Can be scaled to support multiple languages and domains.