Deep Learning Approaches to a School-based Email Classification with TF-IDF-based vectorization

Email has remained the main means of business communication in a school context. It plays a vital role in a school-based communication because it makes it possible for different clients and stakeholders in the educational community to formally communicate effectively, securely, and with documentation, thus the main goal of this research is to classify school-based email data into two categories which is commonly encountered in the records office of a university: General Inquiry Type and Verification Type. The email labeling is done through the use of a rule-based technique that incorporates word embedding, as well as the integration of dictionary-based and user-defined keywords. The study gives emphasis on the use of the dataset of 21,280 emails originating from the researcher's affiliated university. Several deep learning models were employed using Global Vector (GloVe) word embedding and Term Frequency Inverse Document Frequency (TF-IDF) based vectorization. The performance of the deep learning techniques using Word Embeddings on a school-based email data were evaluated based on accuracy, precision, recall, and F1 score. The Convolutional Neural Networks (CNN) model outperformed the other deep learning models which consistently demonstrated the highest results in all metrics used and have a high accuracy of 98% with TF-IDF and 88% with GloVe word embedding. Further, the deep learning models employed are limited only to Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM).