Email Spam Detection using Graph based Convolutional Neural Network with MALO

Every day, millions of individuals and organizations rely on email as a fundamental tool for communication, business transactions, and information exchange. However, the pervasive issue of spam email has escalated in recent years, paralleling the exponential growth in email usage. This surge in spam poses a significant challenge for both individuals and businesses, hindering their ability to effectively manage and filter incoming emails. To confront the complexities inherent in spam email classification, such as the inability of conventional models to adequately capture pertinent information due to document length constraints and the lack of subject matter expertise leading to suboptimal classification outcomes, a novel approach is proposed. This study advocates for the integration of a knowledge graph and a graph convolutional neural network (GCNN) as the foundation for a robust classification model. Utilizing Bidirectional Encoder Representations from Transformers (BERT), the text data is encoded, ensuring a comprehensive representation of semantic meaning. The graph network is then initialized with vectors containing dense semantic information for each word. By leveraging a trained entity-relationship extraction model, the interdependencies between entities and other syntactic information are extracted and utilized as edges within the GCNN. Moreover, the model’s weights are optimized using the principle of Maximum A Posteriori (MAP) in Latent Optimization (MALO). The incorporation of graph structures enables the model to discern intricate semantic relationships between words, enhancing its ability to comprehend the underlying context of the text. This innovative technique transcends traditional email spam detection methods by amalgamating knowledge-driven characteristics with data-driven features, thereby advancing the efficacy of classification endeavors.