Towards robocall and scam call detection

Abstract

Abstract The rapid increase of robocalls and scam calls has become a significant threat to individuals and organizations worldwide, causing financial losses, emotional distress, and breaches of pri- vacy. These malicious calls often use automated systems to impersonate legitimate services or institutions, leading to widespread fraud, identity theft, and financial exploitation. According to estimates, scam calls cost consumers billions of dollars annually, affecting vulnerable populations such as the elderly and those with limited technological awareness. This project focuses on developing an advanced system for the detection and classification of scam and robocalls. We employ a multimodal approach that combines both audio and text analy- sis. For audio modality, we utilize pre-trained models (PTMs) and speech recognition techniques to differentiate between human and automated calls. In the text modality, transformer-based models and large language models (LLMs) are employed to analyze call transcripts and detect linguistic patterns indicative of scams. This research underscores the critical need for automated call classification systems to safeguard privacy, protect against fraud, and mitigate the negative societal impacts of scam calls. The integration of both audio and text-based analysis ensures a robust and scalable solution, paving the way for enhanced security measures in telecommunications. This project contributes to advancing call detection technology, offering a proactive and efficient method for reducing the harm caused by robocalls and scams.