網路拍賣平台詐騙偵測 - 以Yahoo!奇摩為例

Abstract

網路的快速發展，許多創意相應而生，這些創意改變了人類彼此間互動方式，網路拍賣即是一例，網路拍賣平台提供買家與賣家快速交流資訊及交易的機會。然而，成也蕭何、敗也蕭何，許多網路拍賣詐騙犯罪也因此而起。 本論文利用資料探勘中的資料分類法對Yahoo!奇摩網路拍賣平台賣場進行詐騙偵測分析。在取得拍賣平台中的賣場資料後，先針對賣場中的拍賣編號、賣方帳號、開始時間、結束時間、賣場存活時間、所在地區、評價、運費及交貨方式、是否為詐騙、目前出價、出價次數、商品數量、是否通過信用卡認證、是否允許貨到付款、商品新舊、是否接受自取、是否接受面交、是否通過手機認證、評價等欄位進行屬性分析，藉此找出與詐騙關連性大的十一個欄位，分別為賣場存活時間、商品所在城市、賣家評點、是否可自取、是否可面交、是否可貨到付款、商品價格、商品狀態、數量、是否經過信用卡驗證、目前競標人數，再利用資料分類法針對上述十一個欄位進行分析，藉以了解網路拍賣詐騙手法。 本論文以科學的方法找出網路拍賣詐騙之慣用手法，研究所得之Accuracy達97.67%、F-Measure達92.77%，利用資料分類法從資料集歸納所得之詐騙規則如下，當成立的規則愈多則詐騙的可能性愈大： □賣場存活時間小於5天，其中存活時間小於2天，詐騙程度可能性提高； □商品地點於新竹縣、屏東縣、台南縣、台東縣、花蓮縣、高雄縣、基隆市、嘉義縣、苗栗縣、高雄市、台中縣、桃園縣，其中多出現於台東縣、花蓮縣、新竹縣、屏東縣、台南縣； □賣家評價小於44； □允許面交、允許自取但不允許貨到付款； □商品狀態為「全新」或「使用一到三個月」； □賣家帳號通過信用卡驗證； □價格介於3000 ~ 20,000元之間； □詐騙賣場刊登商品日期多集中於下旬。

The rapid advancement of the Internet has led to the development of many innovative ideas. These innovations, in turn, have changed the ways in which human beings interact with each other. Online auction is one such example. It provides buyers and sellers with a platform for quick exchange of information and trading opportunities; unfortunately, many fraudulent scams have also arisen out of these Internet auctions.

This thesis attempts to detect frauds on the Yahoo! Auctions platform by using the data classification technique in Data Mining. Transaction information was firstly gathered from the auction platform. The attributes of a transaction are extracted for analysis, which include item number, seller ID, start and end time for bidding, bidding time remaining, item location, customer review, delivery fee and method, fraud assessment, current bidding price, number of bids, item quantity, credit card authentication, direct payment option, item condition, self-collect option, in-person transaction option, mobile-phone verification, and recent feedback ratings. Attributes with a high-correlation to frauds were identified and further analyzed via data classification. The aim is to provide a better understanding on how online auction frauds are carried out.

This thesis adopted a scientific method to identify common fraudulent practices in Internet auctions. It achieved an Accuracy level of up to 97.67% and an F-Measure of up to 92.77%. From the data sets collected in the study, the following rules with respect to the attributes of a transaction have a prominent correlation with auction frauds. In addition, the more rules that are satisfied, the more possibility that a fraud may occur. □The bidding time is less than five days (risk level was even higher in auctions with bidding time less than two days). □The item is located in in Taitung County, Hualien County, Hsinchu County, Pingtung County, Tainan County, Kaohsiung County, Keelung City, Chiayi County, Miaoli County, Kaohsiung City, Taichung County, Taoyuan County □The transaction permits in-person transaction. □The transaction permits self-collection but not cash against delivery. □The item condition was classified as "new" or "used for one to three months". □The seller has passed the credit card authentication. □The item price is between TWD3000 and 20,000. □The auctions fraud normally occur in the second half of the month