eMailSift: Adapting Graph Mining Techniques for Email Classification

Text classification is the problem of assigning pre-defined class labels to incoming, unclassified documents. The class labels are defined based on a set of examples of pre-classified documents, used as a training corpus. For text classification, a number of approaches have been proposed such as Support Vector machines, Decision trees, k-nearest-neighbor classification, Linear Least Square fit and Bayesian classification among others. The need for handling and classifying large amounts of personal emails have prompted the use of text classification approaches to address email classification. Email classification is trickier than text classification as it is based on personal preferences, consequently it uses disparate criteria which are difficult to quantify. Also, documents are richer in content as compared to emails whose content can vary dramatically from folder to folder, hence conventional approaches may not be well-suited. In addition, as opposed to a static set of corpus typically used for training in text classification, the mail environment is constantly changing, with a need for adaptive and incremental re-training. In this report we propose a novel, graph based mining approach for email classification. Our approach is based on the premise that representative – common and recurring – structures/patterns can be extracted from a pre-classified email folder and the same can be used effectively for classifying incoming email messages. A number of factors that influence representative structure extraction and classification are analyzed conceptually and validated experimentally. In our approach, the notion of inexact graph match is leveraged for deriving structures that provide coverage for characterizing folder contents. Extensive experimentation validate the selection of parameters and the effectiveness of our approach for email classification. We also compare the performance of our approach with the Naive Bayesian classifier.