Distinguishing Sellers Reported as Scammers on Online Illicit Markets Using Their Language Traces

Abstract

Fraud exists on both legitimate e-commerce platforms and illicit dark web marketplaces, impacting both environments. Detecting fraudulent vendors proves challenging, despite clients’ reporting scams to platform administrators and specialised forums. This study introduces a method to differentiate sellers reported as scammers from others by analysing linguistic patterns in their textual traces collected from three distinct cryptomarkets (White House Market, DarkMarket, and Empire Market). It distinguished between potential scammers and reputable sellers based on claims made by Dread forum users. Vendor profiles and product descriptions were then subjected to textometric analysis for raw text and N-gram analysis for pre-processed text. Textual statistics showed no significant differences between profile descriptions and ads, which suggests the need to combine language traces with transactional traces. Textometric indicators, however, were useful in identifying unique ads in which potential scammers used longer, detailed descriptions, including purchase rules and refund policies, to build trust. These indicators aided in choosing relevant documents for qualitative analysis. A pronounced, albeit modest, emphasis on language related to ‘Quality and Price’, ‘Problem Resolution, Communicationand Trust’, and ‘Shipping’ was observed. This supports the hypothesis that scammers may more frequently provide details about transactions and delivery issues. Selective scamming and exit scams may explain the results. Consequently, an analysis of the temporal trajectory of vendors that sheds light on the developmental patterns of their profiles up until their recognition as scammers can be envisaged.

Keywords

• cryptomarket
• fraud
• scammer
• language trace
• forensic linguistic