Skip to Main Content
Product images serve an important role in online auction listings. As thriving businesses, online auction sites often host millions of concurrent auction listings. Where space is limited (such as on the page of auction search results), only product images are displayed to users as an overview of all auction listings. To stand out from competitors, veteran sellers often edit product images to attract potential buyers. Over time, many sellers have developed their own editing styles that recurrently appear in their image pool and are mostly distinct from other sellers, indicating a promising feature for seller profiling. Seller profiling is fundamental for the detection of account anomalies, which are often related to fraudulent acts. Numerous online auction guides suggest that buyers watch for anomalies in a seller's auction listings (such as sudden changes in product categories, auction templates, and text fonts), because such anomalies often indicate account takeovers. Researchers have proposed computational methods to encode such features and automate the detection of anomalies and frauds. However, little previous work has leveraged product images, a major component of auction listings. We developed an automatic algorithm that can extract image editing styles to establish seller profiles.