Social e-commerce tax evasion detection using multi-modal deep neural networks

Social e-commerce is an extension of the hidden economy to the digital realm. Through the convenience of social media platforms as a means of communication, content sharing and even payment between different users, black market transactions are now enabled by e-commerce. Tax authorities worldwide have voiced their concern over the difficulty to detect such transactions over the internet. This paper presents a machine learning based Regtech tool for international tax authorities to detect transaction-based tax evasion activities across social e-commerce. To build such a tool, we collected a dataset of 58,660 Instagram posts and manually labelled 2,041 sampled posts with multiple properties related to transaction-based tax evasion activities. Based on the dataset, we developed a multimodal deep neural network to automatically detect suspicious posts. The proposed model combines comments, hashtags and image modalities (including extracted from videos) to produce the final output. As shown by our experiments, the complementary combined model achieved sensitivity of 71.9 %, specificity of 87.5%, accuracy of 84.1 % and AUC of 0.837, outperforming any single modality models. This tool could help tax authorities to identify audit targets in an efficient and effective manner, and combat social e-commerce tax evasion in scale.