csci544

Naive Bayes classifier to identify hotel reviews as either truthful or deceptive, and either positive or negative.

Procedure

1. Split the given data into training and test set using random sampling.

2. Further, keep a part of training data as development set.

3. Create a Naive-Bayes Classifier using the training data and store the model parameters into a file nbmodel.txt

4. Use the development data to further generate the ouput label for each datapoint.

5. Compare the actual and predicted labels and calulate the Precision, Recall and F-1 score

Run the shell script that automates this pipeline completely.

sh main.sh

Naive-Bayes Model Features and Parameters

Used unigram bag-of-words model as features.

Applied Laplace (Add-1) smoothing to overcome zero probabilities.

Further, Ignored tokens that are not part of intitial training vocabulary.