Results

Baseline

Since five target classes of about equal size are used, the expected score for randomly predicting classes for feature vectors is 0.2.

Evaluations

Below are the results of the evaluations carried out on datasets of different sizes using a multinomial Naive Bayes classifier.
All scores are the average scores of 10 runs of 10-fold stratified cross-validation.

Run 1
Number of feature vectors	608
Vocabulary size	13198
Score	0.484

Run 2
Number of feature vectors	1000
Vocabulary size	18278
Score	0.509