Research On Correlation Networks For Cheating Detection In Distribution Channel

From the perspective of philosophy, the World is a ubiquitous-correlated entire system. The correlation, which is mutual-influenced, mutual-restricted and mutual-interacted, exists between any internal and external factors of things. Given a set of items with their sequential data, correlation networks over these items can be built to represent their associations, and then advanced network analysis methods can be applied on the networks to solve many interesting real-world applications. In this paper, we study how to detect cheating by exploring the correlation networks. As a case study, we focus on the cheating in the distribution channel, through which partners move products from manufacturer to end users. Some research reports show that a lot of world-wide IT companies like Lenovo and IBM gain profit as much as tens of millions of dollars via distribution channel. To increase sales, it is quite common for manufacturers to adjust the product prices to partners according to the product volume per deal. However, the price adjustment is like a double-edged sword. It also spurs some partners to form a cheating alliance, where a cheating seller applies for a falsified big deal with a low price and then re-sells the products to the cheating buyers. Since the cheating behavior is harmful to the healthy operation of distribution channel and results in profit loss of millions of dollars, we need an automatic detecting method to guide the tedious audit process.We observe that if two purchase-volume sequences have strong negative correlation, it is likely that that the abnormal deals happen between these two partners. With this motivation, we develop the framework of cheating detection via correlation network analysis. It mainly addresses the following issues. First, previous correlation measures are usually symmetric and thus cannot distinguish the different roles, namely cheating seller who sells the products from falsified deals to other partners, and cheating buyer who buys the products from other partners, in the abnormal deals. Second, the exhaustive computing of correlation for any pair of sequences may result in some false-positive pairs, whose correlation is purely coincident, and thus should be removed.To this end, first we propose an asymmetric correlation measure to distinguish the roles in the cheating alliance. Also, by assuming that the role of each partner is stable we formulate a new graph cut problem to convert the original network into a bipartite graph, aiming at removing false-positive correlation pairs systematically. Finally, a probabilistic model is proposed to measure the degree of cheating behaviors and generate the ranking of cheating partners. Based on the 4-year channel data of an IT company we empirically show the effectiveness of the proposed method compared with the baseline methods. It is worth mentioning that more than half of the partners in the resultant top-30 ranking list are true cheating partners, and thus this unsupervised method is extremely helpful to cultivate a healthy ecosystem of distribution channel.