What Am I Gonna Say?

🔍 What It Is

• A Shiny app that predicts the next word given a user’s input
• Trained on blogs, news, and Twitter from online datasets
• Uses a stupid backoff n-gram model for fast and simple prediction
• Accepts 1–2 word input and returns the top 3 likely next words

🧠 How It Works

• Trigram model: predicts based on last 2 words
  
• Bigram model: backs off to last 1 word
  
• Unigram fallback: most frequent words
  
• Scoring:
  • Trigrams × 1.0
    
  • Bigrams × 0.4
    
  • Unigrams × 0.16
    
• Returns top 3 candidates by cumulative score

📊 Why It’s Awesome

• Training corpus: 4 million+ lines
  
• Speed: ~50ms average prediction time in Shiny
  
• Accuracy (manual eval):
  • Top-1 match: ~18%
    
  • Top-3 contains correct word: ~37%
    
  • Perplexity (approx.): ~140–180
    
• Model size: <20 MB for deployment
• 💡 Can be enhanced with smoothing, confidence bars, or word embeddings!

🖥️ Live Demo: Shiny App

• Input: how are
  
• Output:
  1. you
     
  2. we
     
  3. they
• Input: let's go
  
• Output:
  1. ahead
     
  2. back
     
  3. now

Simple, fast, intuitive — designed for fun interaction!