2024-12-15
How the Algorithm Works
- Data Preparation: Cleaned, tokenized, and built n-grams (unigrams, bigrams, trigrams).
- Prediction Model:
- Uses the last one to three words of a sentence.
- Searches for the highest-probability n-gram match.
- Handling Unseen Words:
- Backoff model assigns probabilities to unseen n-grams.
- Ensures predictions even for rare word combinations.
App Description
- Purpose: Predict the next word based on user input.
- Features:
- Input box for typing phrases.
- Real-time next-word predictions displayed below.
- Usage:
- Type a partial sentence into the input field.
- See predicted words ranked by likelihood.

Results and Performance
- Accuracy:
- 1st word: 80%
- 2nd word: 70%
- 3rd word: 60%
- Efficiency:
- Average response time: 0.2 seconds.
- Model size: 5MB.
Conclusion and Future Work
- Summary:
- Developed a Shiny app for real-time word prediction.
- Achieved good accuracy and efficiency.
- Future Improvements:
- Train on larger datasets for better predictions.
- Support for additional languages.
- Live App: Shiny App Link