Fighting against online sellers who cheat

Unlike physical shopping malls where visibility is determined by a store’s location, standing out on e-commerce platforms is determined by an item’s ability to rank highly based on a ranking algorithm.

Prominence of position on an e-commerce platform is determined by a ranking system that measures how popular a seller is based on many parameters, including the number of clicks their products receive. To improve their popularity, some sellers may conduct fraudulent transactions where they buy their own products and provide positive feedback through self-owned buyer accounts.

Hoping to curb these types of fake transactions is Assoc Prof Bo An of NTU’s School of Computer Science and Engineering and his collaborators from the Alibaba-NTU Singapore Joint Research Institute and China’s Alibaba Group.

The joint research team developed a novel deep reinforcement learning algorithm and seller behaviour model inferred from real-world data, including both genuine and fake transactions. Their tool improved an e-commerce platform’s impression allocation mechanism by more effectively predicting sellers’ fraudulent behaviour, and outperformed existing fraud-detection algorithms and models.

The research “Impression allocation for combatting fraud in e-commerce via deep reinforcement learning with action norm penalty” was published in Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence (2018), DOI: 10.24963/ijcai.2018/548.
This article appeared first in NTU’s research & innovation magazine Pushing Frontiers (issue #14, December 2018).

You may also like...