This project is mainly utilize Natural Language Processing to predict what a user is going to type. It is mainly depending on the online sources of tons of sentences and phrases. The work is separated into next three steps:

• Access the text data and clean it
• Build and modify the predicting model
• To Present

The text data is from online blogs, news and twitter in English. Because the data is a huge data set. So only 1% of the text data is selected randomly.

Then the profanity, unexpected strings and white spaces are all eliminated. The stop words are kept, Because I found that if stop words are kept, it could increase the precision of prediction.

After cleaning, the text data of sentences are separated into 2-word, 3-word, 4-word, 5-word and 6-word grams.

And then, the words are ranked by frequency, the ones with more frequency would be ranked higher than the ones with lower frequency.

And all the predicted words are kept, no matter whether they are predicted by 2-word, 3-word, 4-word, 5-word or 6-word grams.

But the one predicted by the grams with more words would have priority to the grams with less words.

The prediction model would be presented by the shiny app, which is a user friendly and easy to use online app.

To my surprise, even I loaded a large data set. It responses and calculates fast.

The model I build shows 10 predicted words at most. You could choose the number to show, which is from 1 to 10.

Here is my work ngrams