Introduction

This presentation introduces a predictive text application that suggests the next word a user might type.

The app uses an n-gram language model trained on a large corpus of text from blogs, news articles, and Twitter posts.

Goal: To provide a fast and accurate word prediction experience.

Algorithm

The app uses a statistical language model based on n-grams:

• Quadgrams: Sequences of 4 words
• Trigrams: Sequences of 3 words
• Bigrams: Sequences of 2 words
• Unigrams: Single words

Prediction Logic:

1. The app first looks for matching quadgrams based on the last 3 words entered.
2. If no quadgram is found, it “backs off” to trigrams, then bigrams, and finally unigrams.
3. A discount factor (lambda) is applied during backoff to adjust probabilities.

The App

The Shiny app provides a simple interface:

• Input: A text box where the user enters a phrase.
• Output:
  • The top predicted next word.
  • A table showing the top 3 predictions with their probabilities.

Instructions:

1. Type a phrase in the input box.
2. Click the “Predict” button.
3. View the predicted next word(s).

Link to App: https://ml73o6-mohammed-teslim.shinyapps.io/Teslim_NLP_Project/

How it Works

• The app cleans the input text (removes punctuation, converts to lowercase, etc.).
• It then uses the n-gram model to find the most probable next word based on the preceding words.
• The backoff mechanism ensures that a prediction is always provided, even for unseen word combinations.
• The app is optimized for speed by using efficient data structures (data.table) and pre-computed probabilities.

Conclusion

This app demonstrates the power of n-gram models for next-word prediction.

Future Improvements:

• Implement more advanced smoothing techniques.
• Experiment with larger n-gram models (e.g., 5-grams).
• Explore using recurrent neural networks (RNNs) for improved accuracy.

Thank you!