# Inferring Tracker-Advertiser Relationships in the Online Advertising Ecosystem using Header Bidding

## TL;DR

This paper presents KASHF, a machine learning-based method to infer data sharing relationships between trackers and advertisers in online advertising, including server-side sharing, by analyzing bidding behavior changes.

## Contribution

KASHF introduces a novel approach to detect both client-side and server-side data sharing in online advertising ecosystems through interpretable machine learning.

## Key findings

- Successfully infers relationships regardless of data sharing location
- Identifies server-side data sharing relationships validated externally
- Detects relationships missed by traditional cookie syncing methods

## Abstract

Online advertising relies on trackers and data brokers to show targeted ads to users. To improve targeting, different entities in the intricately interwoven online advertising and tracking ecosystems are incentivized to share information with each other through client-side or server-side mechanisms. Inferring data sharing between entities, especially when it happens at the server-side, is an important and challenging research problem. In this paper, we introduce KASHF: a novel method to infer data sharing relationships between advertisers and trackers by studying how an advertiser's bidding behavior changes as we manipulate the presence of trackers. We operationalize this insight by training an interpretable machine learning model that uses the presence of trackers as features to predict the bidding behavior of an advertiser. By analyzing the machine learning model, we are able to infer relationships between advertisers and trackers irrespective of whether data sharing occurs at the client-side or the server-side. We are also able to identify several server-side data sharing relationships that are validated externally but are not detected by client-side cookie syncing.