• Data Collection:
  • Processed a large text corpus to extract:
    • Unigrams (single words)
    • Bigrams (two-word combinations)
    • Trigrams (three-word combinations)
• Model:
  • Probabilities calculated for n-grams using Add-1 smoothing.
  • Backoff model:
    • Trigrams → Bigrams → Unigrams
• Prediction Flow:
  • User Input → Tokenization → N-Gram Matching → Predicted Word

Model Accuracy by N-Gram

• Trigrams perform best, with 70% accuracy in predictions.
• Bigrams and unigrams serve as fallback mechanisms.

• User experience:
  • Input a sentence → Predict the next word in real-time.
  • Simple and intuitive interface.
• Interactive features:
  • Adjust diversity with sliders.
  • See fallback behavior from trigrams → bigrams → unigrams.

• Accurate Predictions: Context-aware using trigrams and bigrams.
• Diverse Applications: From messaging apps to search engines.
• Customizable: Adjust prediction behavior for specific use cases.
• Scalable: Adapts to new data for continuous improvement.

Ready to revolutionize text input?

Thank you for following this presentation. For more details and to explore the Shiny app, please visit the following link:

https://pietrogazzi.shinyapps.io/DScapstone/