Email Spam Detection using Bidirectional Long Short Term Memory with Convolutional Neural Network

Communication over email in this era of Internet has become very popular on account of its being cheap and easy to use for messaging and sharing important information to others. But spam messages often times make large volume of unwanted messages in the users inbox and it also wastes the resources as well as valuable time of the users. Therefore, in order to identify the message whether it is spam or ham, an efficient and accurate technique is required. In this paper, we propose a new model for detecting spam messages based on the sentiment analysis of the textual data of the email body. We incorporate Word-Embeddings and Bidirectional LSTM network to analyze the sentimental and sequential properties of texts. Furthermore, we speed up the training time and extract higher level text features for Bi-LSTM network using Convolution Neural Network. We involve two datasets namely lingspam dataset and spam text message classification dataset and adopt recall, precision and f-score for comparing and evaluating the performance of our proposed approach. Our model achieves improved performance of accuracy about 98-99%. Apart from this, we demonstrate our model outperforms not only to some popular machine learning classifiers but also to state of the art approaches for detecting spam messages and hence, proves the superiority by itself.