Word prediction by n-gram model

This prediction model takes a text string, a decay factor f (between 0 and 1) and a number top_n as input, and then prints the top n predictions following the string.

• Example input: predict(“i think”,f=1,top_n=5)
• Output: “i” “its” “so” “that” “about”

The n-grams used for prediction were collected from a dataset of blogs, tweets and news. A set of 1, 2, 3, 4 and 5-grams were generated from the training set.

• The n-grams used for prediction were collected from a dataset of blogs, tweets and news. A set of 1, 2, 3, 4 and 5-grams were generated from the training set.
• A subset containing 25% of the data was sued for training
• All characters converted to lowercase, numbers and puncutuation removes

Given a string, the function finds

• the last four words of the string
• finds commonest 5-grams beginning with the string with their frequencies
• applies the function recursively to last 3, 2 and 1-grams of the string in sequence
• accumulates the frequencies of commonest predicted words in a dataframe
• the variable f (between 0 and 1) is the decay factor; for every recursive call to the function, the word counts are multiplied by fxf (i.e. the relative importance of words predicted diminishes)

Probable words follwing 'i think'

The program runs with only moderate memory usage (19.7 MB saved environment data), which makes it useful for smartphones. However, predictive accuracy is yet to be tested in real world.