Best Practices for Community and E‑commerce Fraud Prevention on Xiaohongshu: Understanding and Combating Fake Traffic

This article focuses on the best practices of community and e‑commerce risk control at Xiaohongshu, specifically the governance of fake traffic (刷量).

As a content community, Xiaohongshu’s monthly active users surpassed 200 million in October 2021 and continue to grow, making it a target for fake‑traffic operations.

From a user perspective, security must ensure (1) the safety of user information online and (2) the authenticity of the information users obtain.

Fake traffic, a common cheating method in both community and commerce, creates false metrics such as likes, comments, views, and sales, undermining content safety and user trust.

1.1 What is fake traffic? It is data falsification—using cheating methods to generate artificial likes, collections, etc., for higher commercial value. In e‑commerce, it often involves hiring fake buyers and fabricating logistics to inflate GMV.

The article examines three types of fake‑traffic service companies:

1.1.1 Group‑controlled fake‑traffic companies use jail‑broken or normal iPhones with Apple’s image‑packaging mechanism to clone login states and run batch sandbox environments for large‑scale operations.

1.1.2 Account‑nurturing (养号) companies register virtual accounts, sell high‑imitation clothing, and after accounts are blacklisted, swap them for cheap phones, then employ real people to continuously publish content and drive brand traffic.

1.1.3 Crowdsourced fake‑traffic companies recruit part‑time workers on major platforms to complete tasks with their own accounts for payment.

1.2 Redefining fake traffic It is likened to counterfeit currency in a market economy: likes, reads, and comments are the “currency” of the community, and fake traffic inflates this currency, leading to market inflation.

2.1 Identification challenges include terminal uncontrollability, AI‑vs‑AI, and human‑vs‑human. Terminals are hard to control; attackers can reverse‑engineer protocols and spoof signatures. AI models face “AI‑vs‑AI” arms races, and real‑person crowdsourced attacks are difficult to detect.

2.2 Governance challenges stem from multiple stakeholder roles (K‑bloggers, brands, MCNs, ordinary users) all potentially participating in fake traffic, making attribution and avoiding collateral damage complex.

2.3 Consolidation challenges involve measuring ROI of anti‑fraud investments and evaluating effectiveness beyond simple CTR metrics.

3 Solution directions

3.1 Risk‑control infrastructure engineization – building an engine that connects the entire upstream and downstream pipeline, enabling near‑real‑time rule assembly, multi‑source data integration, behavior‑sequence analysis, graph analysis, and model‑driven decisions.

Illustrations of the near‑line system architecture and its integration with the unified business gateway (edith2.0) are provided.

3.2 Process‑oriented risk‑control – a five‑step workflow: risk perception, capability building, risk identification, risk mitigation, and effect evaluation. This standardizes analysis, feedback, and iterative improvement.

3.3 Intelligent risk identification – three stages:

Behavior‑entity feature analysis (machine‑level low‑cost attacks).

Group‑level feature mining (clustering devices to catch coordinated attacks).

Graph‑based entity discovery (detecting human‑level coordinated fake traffic via community and commerce graphs).

Feature engineering includes selecting device, account, and interaction attributes, standardizing them, and feeding them into regression models for scoring.

Clustering models use device specifications to group similar devices, creating an unsupervised high‑recall, high‑precision blacklist.

Graph algorithms propagate fraud labels from seed users through device, third‑party account, and content relationships to uncover hidden malicious actors.

Finally, a three‑point methodology is proposed: eliminate the impact of fraud, target the perpetrators, and remove the motivation by combining technical detection with legal and operational actions. Since 2022, Xiaohongshu has removed 31 billion fraudulent likes, demonstrating a zero‑tolerance stance.