Classification of english phrases and SMS text messages using Bayes and Support Vector Machine classifiers

This paper performs a comparative analysis of several different types of SMS text classifiers: weight enhanced Multinomial naive Bayes, Poisson naive Bayes, and L2-loss Support Vector Machine. The effects of preprocessing and incorporating additional features on the classifiers were examined. The preliminary experimental results show that the use of preprocessing and incorporating additional features produced no significant gain or loss in classification efficiency. However the feature space used by the classification methods decreased, which could be beneficial for resource limited environments. In addition the solutions to the SMS text classification may be applied to other problems, like the classification of English sentences. Our collection of text messages may not be statistically significant, because of very limited sources for text messages.