Build a N-gram model to predict a single word

The purpose of our project is to build a predictive model to predict a single word given a previous words or phase entered as input.

The objectifs needed to achieve the purpose are:

• Get a training data sets ;
• Clean the data ;
• Build a n-gram terms with training data ;
• Use appropriate algorithm to build a predictive model with n-gram terms.

• We collected training dataset from publicly available sources such as newspaper, personal blog and Tweeter ;

• We selected a random sample from training data sets ;

• We cleaned the data with R tm package and build a 1-gram, 2-gram and 3-gram data.frames ;

• We applied Maximum likelihood algorithm to compute individual n-gram probablities and selected word with the highest probability.

The model we built use 2 inputs to produce 2 outputs:

The inputs are:

• Document type giving the background of prediction (blog, news or twitter) ;
• The phrase or set of words that is provided.

The ouputs are:

• The next word predicted given the phrase you entered ;
• The perplexity that measure the quality of the prediction.

• Select the document type (blogs, news or twitter) ;
• Enter an English phrase in the textbox ;
• Click on submit button to predict the next word ;
• The model will preprocessed the phrase to remove white spaces, stop words, punctuations, number, etc.;
• If the last two words of are in 1-gram vocabulary, the prediction will give a word ;
• If any of the two words are in 1-gram vocabulary, the prediction will give NONE and unknown for perplexity ;
• The best prediction has perplexity = 1 and the worse has perplexity = Inf!