Multilayered Perceptron Feed-Forward Artificial Neural Network Approach for E-Mail Classification
Ogwueleka Francisca Nonyelum · SSRN Electronic Journal · 2012
E-mail messages are originally designed to be sent and accumulated in repository for periodical use which amounts to the details of an event or a meeting’s upcoming agenda for a particular organization. These messages range from static organizational knowledge to conversations and pose a lot of difficulties to users in terms of prioritizing and processing of the contents of both stored and new incoming messages. This research has established a classification model which classifies the accumulated e-mails in the mail inbox known as dataset into four classes: critical, urgent, important and others. Electronic mail extractor application was designed and implanted to extract e-mail contents. The application used heuristic technique based on Term Frequency-Inverse Document Frequency (TF-IDF) to determine what keywords in a dataset of e-mail messages might be more favorable to use in a query. Nuclass 7.1 Artificial Neural Network (ANN) software was used in the implementation of automated e-mail classification into user defined word classes corresponding to pre-formated class identity, and was able to learn in an associative learning approach, in which the network was trained by providing it with input and matching output patterns. This study showed that Neural Networks (NN) using Back Propagation (BP) technique combined with electronic mail extractor can be successfully used for automated e-mail classification into meaningful classes.