Detecting Crowdsourcing Click Fraud in Search Advertising Based on Clustering Analysis
Jiarui Xu, Li Chen · 2015
This article analyzes the denseness, moderateness and concentricity of search advertising crowd sourcing click fraud. Base on this, it puts forward the click fraud detection model based on clustering analysis. The model includes three steps: preprocessing, group detecting and post-processing. In the preprocessing step, the query that is less likely to be fraudulently clicked is removed. In the group detection step, a crowd sourcing click fraud group is equivalent to a cluster. DP-Means clustering method is used to detect malicious groups. In the post-processing step, demand clicks checked by mistake are filtered. The convergence, scalability and accuracy are verified by the simulation data and one week click data of a search engine company.