CONTENT-BASED SCAM CALL IDENTIFICATION USING SPARSE ATTENTION NEURAL NETWORKS

Abstract

The increasing prevalence of scam calls has become a major cybersecurity concern, causing financial losses and privacy breaches worldwide. Traditional spam detection systems rely heavily on static rule-based filtering or metadata analysis, which often fails to detect sophisticated scam calls that use dynamic conversational patterns. This research proposes a novel deep learning-based framework for classifying scam calls through content analysis using Dynamic Sparsity Top-K Attention Regularization. The model leverages natural language processing (NLP) techniques to analyze call transcripts and identify malicious intent based on semantic patterns, linguistic cues, and contextual dependencies. Unlike conventional attention mechanisms that process all tokens equally, the proposed approach introduces a Top-K attention regularization strategy that dynamically selects the most relevant features, thereby reducing computational complexity and improving classification accuracy. The dynamic sparsity mechanism ensures that only the most informative tokens contribute to the final decision, making the model efficient and suitable for realtime applications. Additionally, the system integrates speech-to-text conversion, feature extraction, and attention-based neural networks to provide end-to-end scam detection. Experimental results demonstrate that the proposed model achieves higher precision, recall, and F1-score compared to traditional machine learning and baseline deep learning approaches. Furthermore, the model shows reduced inference time and improved scalability, making it suitable for deployment in telecom systems and mobile applications. The proposed framework enhances user security by providing real-time scam detection and contributes to the development of intelligent communication systems capable of identifying fraudulent activities. Overall, this research highlights the effectiveness of combining NLP, attention mechanisms, and sparsity optimization techniques in detecting scam calls and improving communication security in modern telecommunication networks