Next Word Predictor App

Project Overview

• Developed a web-based Next Word Predictor App using R and Shiny.
• Built on Natural Language Processing and n-gram frequency models.
• Data source: SwiftKey corpus (blogs, news, Twitter).
• Live at: shinyapps.io

---

How the Model Works

• Corpus loaded and cleaned with tm and stringr packages.
• Created unigrams, bigrams, trigrams using tidytext.
• Prediction logic:
  • 3+ word input → trigram match
  • 2 words → bigram match
  • 1 word or no match → unigram fallback
• Fast lookup using frequency tables saved as .RDS

---

Live App Demo

• User interface built in Shiny with:
  • Text input box
  • Output showing predicted next word
• Hosted at:
  https://chandancsingh52.shinyapps.io/nextwordpredictorapp
• Simple, lightweight & easy to use

---

Performance & Testing

• Preprocessed >2 million lines of text
• Response time: < 1 second
• Tested on real English phrases from Twitter/news
• Output reasonable predictions in >90% of test cases
• Optimized for speed over deep accuracy

---

Conclusion & Future Work

• Great proof-of-concept of applied NLP in R
• Can be expanded to:
  • Suggest multiple predictions
  • Visualizations (word cloud, frequency charts)
  • Advanced models (LSTM, BERT)
• A strong foundation for AI-powered typing tools
• Ready for production — just needs more data 🚀