J48S: A Sequence Classification Approach to Text Analysis Based on Decision Trees

Sequences play a major role in the extraction of information from data. As an example, in business intelligence, they can be used to track the evolution of customer behaviors over time or to model relevant relationships. In this paper, we focus our attention on the domain of contact centers, where sequential data typically take the form of oral or written interactions, and word sequences often play a major role in text classification, and we investigate the connections between sequential data and text mining techniques. The main contribution of the paper is a new machine learning algorithm, called J48S, that associates semantic knowledge with telephone conversations. The proposed solution is based on the well-known C4.5 decision tree learner, and it is natively able to mix static, that is, numeric or categorical, data and sequential ones, such as texts, for classification purposes. The algorithm, evaluated in a real business setting, is shown to provide competitive classification performances compared with classical approaches, while generating highly interpretable models and effectively reducing the data preparation effort.