# A Marketplace Price Anomaly Detection System at Scale

**Authors:** Akshit Sarpal, Qiwen Kang, Fangping Huang, Yang Song, Lijie Wan  
**arXiv:** [2310.04367](https://arxiv.org/abs/2310.04367)  
**Date:** 2023-10-06  
**Summary page:** https://www.summarizepaper.com/en/paper/2310.04367v2/

## Key points

- Authors address challenges faced by online marketplaces in managing a large volume of price updates initiated by individual sellers
- Democratization of pricing on platforms leads to an increase in data quality issues
- Lack of centralized guardrails can result in inaccurate prices being published, impacting customer experience and posing a risk of revenue loss
- Introduction of MoatPlus, a scalable price anomaly detection framework for growing marketplace platforms
- Objective of MoatPlus is to leverage proximity and historical price trends to establish an upper price bound using unsupervised statistical features
- Framework employs ensemble models to detect irregularities, exclude problematic elements, and utilize optimized weighting scheme for reliable price boundary within real-time pricing pipeline
- Significant improvements observed in precise anchor coverage, especially in high-vulnerability item subsets
- Innovative approach enhances accuracy and reliability of pricing while mitigating risks associated with erroneous pricing strategies
- Implementation of MoatPlus offers promising solution for marketplace operators to maintain data integrity and optimize pricing strategies at scale

## Summary

In their paper titled "A Marketplace Price Anomaly Detection System at Scale," authors Akshit Sarpal, Qiwen Kang, Fangping Huang, Yang Song, and Lijie Wan address the challenges faced by online marketplaces in managing a large volume of price updates initiated by individual sellers. The democratization of pricing on these platforms has led to an increase in data quality issues. The lack of centralized guardrails present in traditional online retail settings can result in inaccurate prices being published. This not only impacts customer experience but also poses a risk of revenue loss for marketplace operators. To tackle this issue, the authors introduce MoatPlus, a scalable price anomaly detection framework designed specifically for growing marketplace platforms. The key objective of MoatPlus is to leverage proximity and historical price trends derived from unsupervised statistical features to establish an upper price bound. By employing an ensemble of models, the framework is able to detect irregularities in price-based features, exclude problematic elements, and utilize an optimized weighting scheme to construct a reliable price boundary within the real-time pricing pipeline. Through their research, the authors observed significant improvements in precise anchor coverage, particularly in high-vulnerability item subsets. This innovative approach not only enhances the accuracy and reliability of pricing on online marketplaces but also contributes towards mitigating potential risks associated with erroneous pricing strategies. Overall, the implementation of MoatPlus offers a promising solution for marketplace operators seeking to maintain data integrity and optimize pricing strategies at scale.