【ICDM 2022教程】图挖掘中的公平性:度量、算法和应用

来源：专知
本文为书籍介绍，建议阅读5分钟
本教程全面概述了在测量和减轻图挖掘算法中出现的偏差方面的最新研究进展。

Part 1: 引言Introduction

• Background and Motivation.
• An overview of graph mining tasks that have been studied on algorithmic bias mitigation.
• An overview of the applications which benefit from debiased graph mining algorithms.

Part 2:图挖掘公平性符号与度量 Fairness Notions and Metrics in Graph Mining

• Why is it necessary to define fairness in different ways?
• Group Fairness: graph mining algorithms should not render discriminatory predictions or decisions against individuals from any specific sensitive subgroup.
• Individual Fairness: graph mining algorithms should render similar predictions for similar individuals.
• Counterfactual Fairness: an individual should receive similar predictions when his/her features are perturbed in a counterfactual manner.
• Degree-Related Fairness: nodes with different degree values in the graph should receive similar quality of predictions.
• Application-Specific Fairness: fairness notions defined in specific real-world applications.

Part 3: 图挖掘算法去偏见技术 Techniques to Debias Graph Mining Algorithms

• Optimization with regularization.
• Optimization with constraint.
• Adversarial learning.
• Edge re-wiring.
• Re-balancing.
• Orthogonal projection.

Part 4: 真实世界应用场景 Real-World Application Scenarios

• Recommender systems.
• Applications based on knowledge graphs.
• Other real-world applications, including candidate-job matching, criminal justice, transportation optimization, credit default prediction, etc.

Part 5: 总结 挑战与未来 Summary, Challenges, and Future Directions

• Summary of presented fairness notions, metrics and debiasing techniques in graph mining.
• Summary on current challenges and future directions.
• Discussion with audience on which fairness notion, metric should be applied to their own application scenarios.