Using Data Mining Techniques and the Choice of Mode of Text Representation for Improving the Detection and Filtering of Spam
Reda Mohamed Hamou, Abdelmalek Amine · Advances in business information systems and analytics book series · 2015
This chapter studies a boosting algorithm based, first, on Bayesian filters that work by establishing a correlation between the presence of certain elements in a message and the fact that they appear in general unsolicited messages (spam) or in legitimate email (ham) to calculate the probability that the message is spam and, second, on an unsupervised learning algorithm: in this case the K-means. A probabilistic technique is used to weight the terms of the matrix term-category, and K-means are used to filter the two classes (spam and ham). To determine the sensitive parameters that improve the classifications, the authors study the content of the messages by using a representation of messages by the n-gram words and characters independent of languages to later decide what representation ought to get a good classification. The work was validated by several validation measures based on recall and precision.