ScamGen: Unveiling psychological patterns in tele-scam through advanced template-augmented corpus generation

Xu Han, Qiang Li, Yaling Qi, Hongbo Cao, Witold Pedrycz, Wei Wang

Abstract: Highlights•Introduces ScamGen, a technique for enhancing Chinese telephone scam data.•Leverages psychological insights to generate diverse scam scenarios.•Expands a seed dataset using sentence- and word-level perturbations.•Outperforms large language models in quality and variety of datasets.•Develops deep learning models for intent detection, with BERT achieving 86.68%.