This presentation is created as a part of the Capstone project in the Data Science Course on Coursera by John Hopkins university. The source of the data used in building the model is available in coursera which in turn has been provided by the course partner SwiftKey.

The project features an algorithm for predicting next word in a sentence. The outcome of the project is a Shiny App which provides an UI interface to interact with the algorithm.

The data source consists of text in English language sourced from blogs, news articles and X (twitter) feeds. The data is organized as sentences in the first phase and then sampled to work with reduced amount of data. For this project 1% and 5% sampling was tested. At the end 1% sampling has been used to ensure that app can be hosted online due to memory constraints.

Some form of cleaning is done on the sampled data to remove, numbers, punctuation marks and special characters. Then a N-Gram analysis using ngram_asweka from ngram library is applied to generate tokens of words along with frequencies. These are stored in RData files. Based on the words at prompt, a look-up operation is carried out with the saved N-Gram data sets to determine the most likely next word.

The source code for the application Next Word Predictor can be found at: https://github.com/hisscaredbrain/NextWordPredictor

The application itself is hosted on the shinyapp.io portal under the link : https://curiousity.shinyapps.io/NextWordPrediction/

Enter the text at the prompt on the left hand panel and hit the button Submit. The best three predictions shall be display in the main panel.

The data cleaning is not effective as evidenced with presence of ordinal indicators such as th, st, or rd etc. Due to the low sampling there are instances where second and third predictions are not available when input prompt with 4 or 5 words.

This marks the end of my data science learning in Coursera.

Thank you for reviewing it and helping me learn.