Explaining recommender systems fairness and accuracy through the lens of data characteristics
作者：
Highlights：
• Systematic analysis of impact of data characteristics on recommendation performance.
• Regression-based explanatory modeling targeting accuracy and fairness metrics.
• A minimum set of characteristics with maximum explanatory power is derived.
• The amount of explainability achieved is larger for accuracy than fairness.
• Injecting data biases in the model allows to improve explanatory power for fairness.
摘要
•Systematic analysis of impact of data characteristics on recommendation performance.•Regression-based explanatory modeling targeting accuracy and fairness metrics.•A minimum set of characteristics with maximum explanatory power is derived.•The amount of explainability achieved is larger for accuracy than fairness.•Injecting data biases in the model allows to improve explanatory power for fairness.
论文关键词：Explanatory power,Fairness,Accuracy,Collaborative filtering,Data characteristics
论文评审过程：Received 21 December 2020, Revised 21 April 2021, Accepted 15 June 2021, Available online 1 July 2021, Version of Record 1 July 2021.
论文官网地址：https://doi.org/10.1016/j.ipm.2021.102662
