SCAM DETECTION IN ONLINE CLASSIFIED ADVERTISEMENTS

Abstract

Online classified ad websites have become one of the most fundamental parts of the advertisement industry. Popular customer-to-customer marketplaces such as Craigslist and eBay, have attracted millions of consumers for trading and purchasing secondhand items. Because of the high financial return sellers can gain by using these sites and the anonymity some websites provide by not requiring their users to create user accounts to post ads, online classified sites have a high potential for fraudulent activities. The primary objective of this theses is to develop a computational approach to scam detection in online classified ads. In this research, we first highlight the unique characteristics of scams compared with spam and provide a definition of scams in the general context of online classified ads; then we identify a set of novel features that signal scam or legitimate ads based on the heuristics we derived from observing and exploring the real-world data; and finally, we develop machine learning models for detecting scams in online classified ads and test the models with real-world data collected from Craigslist. The experiment results show that the proposed scam detection models achieved an F-measure of 0.955. The findings of this theses have significant implications for improving the trustworthiness of customer-to-customer online marketplaces.