A Method of Mining Key Accounts from Internet Pyramid Selling Data

Abstract

Internet pyramid selling causes great harm and has difficulty in obtaining evidence. As most of internet pyramid selling often uses virtual coins to complete cash settlement, criminals can exchange virtual coins by controlling visual accounts to transfer funds illegally. The purpose of the work is to mine the key accounts and build evidence chain of personnel and funds. In order to deal with a large number of transaction data, we characterize the virtual currency trading data, which is divided into seven transaction characteristics. The key accounts are these outliers which have too prominent trading behaviour. So we use positive samples to train One Class Support Vector Machine (OCSVM) classification to find trading behaviours' boundary and detect outliers in classification. Then, the correlation between member's information and pyramid selling hierarchical relationship can also be obtained by further linking the trading behaviour of key accounts. Finally, we can screen out the virtual accounts directly controlled by the pyramid selling organization, analyse the flow of funds, investigate and obtain evidence on amount of money involved in pyramid selling. The experimental results show that the classification model cannot only establish the normal model for account transfer behaviour, but also effectively identify abnormal transfer behaviour, thus improving efficiency of investigation and evidence collection for economic investigation departments.