An Assessment of Case Base Reasoning for Short Text Message Classification

Abstract

Message classification is a text classification task that has provoked much interest in machine learning. One aspect of message classification that presents a particular challenge is the classification of short text messages. This paper presents an assessment of applying a casebased reasoning approach that was developed for long text messages (specifically spam filtering) to short text messages. The evaluation involves determining the most appropriate feature types and feature representation for short text messages and then comparing the performance of the case-based classifier with both a Naive Bayes classifier and a Support Vector Machine. Our evaluation shows that short text messages require different features and even different classifiers than long text messages. A machine learner which is to classify text messages will require some level of configuration in these aspects.