Jiaming Huang’s research while affiliated with Alibaba Group and other places

Graph-powered fraud detection is a common issue in various areas, such as e-commerce, banking, insurance and social networks, where data can be naturally formulated as graph structure. Especially in e-commerce, due to its large scale and enormous amount of real-time transactions over millions of merchandises, fraud detection has become an important and serious problem. The challenges lie in three aspects: sparse fraud samples, complex features in online transactions and extra-large scale of e-commerce data. To deal with above issues, in this paper, we propose an efficient graph-powered large-scale fraud detection framework. Concretely, we first present a heterogeneous label propagation algorithm to recall more potentially fraudulent samples for further model training; then, we design a novel multi-view heterogeneous graph neural network model to obtain more accurate fraud predictions; finally, a fraud pattern analysis approach is presented to discover hidden fraud groups. In addition, in order to improve the efficiency and scalability of our proposed fraud detection framework, we present a large-scale fraud detection system deployed on a general graph computing engine. We conduct experiments on two real-world datasets. Results show that the proposed graph-powered fraud detection framework achieves high accuracy and superior scalability on large-scale graph data.

In e-commerce scenarios, frauds events such as telecom fraud, insurance fraud, and fraudulent transactions, bring a huge amount of loss to merchants or users. Identification of fraudsters helps regulators take measures for targeted control. Given a set of fraudsters and suspicious users observed from victims’ reports, how can we effectively distinguish risky users closely related to them from the others for further investigation by human experts? Fraudsters take camouflage actions to hide from being discovered; complex features on users are hard to deal with; patterns of fraudsters are sometimes difficult to explain by human knowledge; and real-world applications involve millions of users. All this makes the question hard to answer. To this end, we design eRiskCom, an e-commerce risky community detection platform to detect risky groups containing identified fraudsters and other closely related users. With the hypothesis that users who interact frequently with fraudsters are more likely to come from the same “risky community,” we construct a connected graph expanded from the identified fraudsters and suspicious users. Next, graph partition is employed to get knowledge of assignment of identified users to potential risky communities, followed by pruning to discover the core members of each community. Finally, top-K users with a high risk score in the neighborhood of core members of each potential community form a final risky community. The extensive experiments are conducted to analyze the effect of our platform components on the alignment with requirements of practical scenarios, and experimental results further demonstrate that eRiskCom is effective and easy to deploy for real-world applications.

With the development of e-commerce, fraud behaviors have been becoming one of the biggest threats to the e-commerce business. Fraud behaviors seriously damage the ranking system of e-commerce platforms and adversely influence the shopping experience of users. It is of great practical value to detect fraud behaviors on e-commerce platforms. However, the task is non-trivial, since the adversarial action taken by fraudsters. Existing fraud detection systems used in the e-commerce industry easily suffer from performance decay and can not adapt to the upgrade of fraud patterns, as they take already known fraud behaviors as supervision information to detect other suspicious behaviors. In this article, we propose a competitive graph neural networks (CGNN)-based fraud detection system (eFraudCom) to detect fraud behaviors at one of the largest e-commerce platforms, “Taobao” ¹ . In the eFraudCom system, (1) the competitive graph neural networks (CGNN) as the core part of eFraudCom can classify behaviors of users directly by modeling the distributions of normal and fraud behaviors separately; (2) some normal behaviors will be utilized as weak supervision information to guide the CGNN to build the profile for normal behaviors that are more stable than fraud behaviors. The algorithm dependency on fraud behaviors will be eliminated, which enables eFraudCom to detect fraud behaviors in presence of the new fraud patterns; (3) the mutual information regularization term can maximize the separability between normal and fraud behaviors to further improve CGNN. eFraudCom is implemented into a prototype system and the performance of the system is evaluated by extensive experiments. The experiments on two Taobao and two public datasets demonstrate that the proposed deep framework CGNN is superior to other baselines in detecting fraud behaviors. A case study on Taobao datasets verifies that CGNN is still robust when the fraud patterns have been upgraded.

Graph computing is more and more widely used in various fields such as spatial information network and social network. However, the existing graph computing systems have some problems like complex programming and steep learning curve. This paper introduces GRAPE, a distributed large-scale GRAPh Engine, which has the unique features of solid theoretical guarantee, ease of use, auto-parallelization and high performance. The paper also introduces several typical scenarios of graph computing, including entity resolution, link prediction, community detection and graph mining of spatial information network. In these scenarios, various problems have been encountered in the existing systems, such as failure to compute over large-scale data due to the high computation complexity, loss of accuracy due to the cropping of original data and too long execution time. In the face of these challenges, GRAPE is easy to support these computing scenarios with a series of technical improvements. With the deployment of GRAPE in Alibaba, both effectiveness and efficiency of graph computing have been greatly improved.