Next Word Prediction

author: VK date: September 2026 autosize: true

Data Science Capstone Project

A simple NLP application that predicts the next word from a phrase.

Based on English blogs, news and Twitter text.

The application uses an n-gram based prediction approach.

• The text data is cleaned and converted into words.
• Common bigrams and trigrams are extracted.
• The most frequent word sequences are stored in the model.
• When a user enters a phrase, the last words are used for prediction.
• Trigram matching is attempted first, followed by lower-order sequences when needed.

The model balances prediction quality with speed and memory usage.

The exploratory analysis used a sample of 30,000 lines from the three English data sources.

Key quantitative results:

• Total tokens analyzed: 879,665
• Unique words covering about 50% of the text: 139
• Unique words covering about 90% of the text: 7,067
• Top 10 bigrams and trigrams were identified from the sample.
• The final model stores the most frequent 10,000 unigram words.

For the final application test, 5 sample phrases were entered and the application produced a single-word prediction for all 5 phrases.

The final product is a Shiny web application.

How it works:

1. Enter a phrase in the text box.
2. Click “Predict Next Word”.
3. The application analyzes the entered phrase.
4. The model searches for matching word sequences.
5. A single predicted next word is displayed.

The application is deployed on shinyapps.io and can be accessed through a web browser.

The project demonstrates a practical NLP next-word prediction system.

Main outcomes:

• English text from blogs, news and Twitter was analyzed.
• Word, bigram and trigram frequencies were explored.
• An n-gram prediction model was developed.
• A Shiny interface was created for real-time prediction.
• The application successfully produced predictions for test phrases.

The project focuses on a simple, fast and usable prediction experience.