Project Description
Together with SingularityNET, I helped develop a weighted liquid rank reputation algorithm which aims at reducing recommendation fraud via experiments on simulated Amazon-like marketplaces.
At the scope of this work I coauthored several papers, which are listed here. For a simpler explanation of what the algorithm does, one can follow this blog post.
Papers:
A Liquid Democracy System for Human-Computer Societies
A Reputation System for Multi-Agent Marketplaces
A Reputation System for Marketplaces - Viability Assessment
Research is still ongoing, the reputation system is also about to be practically used on marketplaces. As argued in a blog post, this system is performing exceptionally well agains recommendation fraud. That is - it prevents gaming the system - most sellers buy fake reviews/ratings in order to give better ratings to their products. This indicates to recommendation system that this product is good, recommending it to more people who then purchase suboptimal goods. According to this analysis (which was part of my work) roughly 10% of the reviews and ratings on Amazon marketplace are fake.