logseq/pages/RecSys_2024_paper_810.md

type:: REVIEWS tags:: year:: 2024 venue:: RecSys full-title:: Understanding Fairness Metrics in Recommender Systems: A Healthcare Perspective date-start:: 17-08-2024 - 07:07 date-submitted:: external-links:: status:: done deadline-submission:: file:: parent:: todoist::

- ### [[Highlights]]
	-
	- # Annotazioni
		- (17/8/2024, 07:03:27)
		- - “four fairness metrics – Demographic Parity, Equal Accuracy, Equalized Odds, and Positive Predictive Value” ([“RecSys_2024_paper_810”, p. 1](zotero://select/library/items/IYHR8UGG)) ([pdf](zotero://open-pdf/library/items/QTBQ68HB?page=1&annotation=3GTNYIC9)) #00b036
		- - “need for enhanced information and education on algorithmic fairness to support informed decision-making in using these systems.” ([“RecSys_2024_paper_810”, p. 1](zotero://select/library/items/IYHR8UGG)) ([pdf](zotero://open-pdf/library/items/QTBQ68HB?page=1&annotation=5D7EJ2FJ)) #00b036
		- - “Algorithmic fairness, in particular, has become a major issue, especially when these systems affect individuals’ lives and well-being” ([“RecSys_2024_paper_810”, p. 1](zotero://select/library/items/IYHR8UGG)) ([pdf](zotero://open-pdf/library/items/QTBQ68HB?page=1&annotation=S68EC3UD)) #00b036
		- - “Demographic Parity focuses on equal representation, while Equalized Odds ensures equal error rates across groups.” ([“RecSys_2024_paper_810”, p. 1](zotero://select/library/items/IYHR8UGG)) ([pdf](zotero://open-pdf/library/items/QTBQ68HB?page=1&annotation=GHKF6A7E)) #00b036
		- - “This divergence underscores a fundamental challenge in designing recommender systems that are perceived as fair by all users.” ([“RecSys_2024_paper_810”, p. 3](zotero://select/library/items/IYHR8UGG)) ([pdf](zotero://open-pdf/library/items/QTBQ68HB?page=3&annotation=MK3QGMZQ)) #f0ff00
		- *Is it really possible? Does it make sense?*
		- - “This disparity in perceptions also points to a broader issue in the deployment of recommender systems: the communication and understanding of algorithmic decisions.” ([“RecSys_2024_paper_810”, p. 4](zotero://select/library/items/IYHR8UGG)) ([pdf](zotero://open-pdf/library/items/QTBQ68HB?page=4&annotation=4MTE8WGC)) #f0ff00
		- *It can be also the conte t and the gravity. The question is if metrics alone are enough or we need something note sophisticated that for instance can choose metrics that make sense with respect to the context.*
		- - “explanations of how algorithms function and the trade-offs they entail. Users should understand not only the mechanics of the algorithms but also the ethical considerations involved in choosing one fairness metric over another. Furthermore, the” ([“RecSys_2024_paper_810”, p. 4](zotero://select/library/items/IYHR8UGG)) ([pdf](zotero://open-pdf/library/items/QTBQ68HB?page=4&annotation=ZJ7YPMXZ)) #f0ff00
		- *Yes exactly.*
		- - “further complicating the task of developing universally fair systems. The apparent lack of a universally” ([“RecSys_2024_paper_810”, p. 4](zotero://select/library/items/IYHR8UGG)) ([pdf](zotero://open-pdf/library/items/QTBQ68HB?page=4&annotation=6CNLB6B4)) #f0ff00
		- *I'm not sure this can be done.*
- ### [[Comments]]
	- The paper concerns the critical issue of fairness in AI-driven decision-making systems, specifically within the healthcare sector. It provides an exploration of how different fairness metrics are understood by the public and their implications in healthcare scenarios. The study highlights fairness's complexity and contextual nature, emphasizing the need for educating people on algorithmic fairness to follow informed decision-making processes.
	- ### Comments:
		- The paper is about a relevant and timely issue in AI ethics. I am positive about the topic and the paper's approach overall. I recommend some minor revisions to expand the paper on some critical aspects as discussed below:
			- 1.  The paper discusses four fairness metrics, each with distinct implications. However, the paper could benefit from further elaboration on the trade-offs involved when applying these metrics (or even additional ones) in different contexts, as noted on page 3 of the paper: "This divergence underscores a fundamental challenge…").
			- 2.  The question of whether metrics alone are sufficient or if a more sophisticated, context-aware approach is necessary is crucial. The paper aligns with this by advocating for better communication and education on the ethical considerations of algorithmic decisions. Discussing the research and technical challenges to achieve such a goal would strengthen the paper.
			- 3.  I'm skeptical about the feasibility of creating universally fair systems. The paper could benefit from a more grounded discussion on the practical limitations of implementing fairness in real-world healthcare and general settings, possibly discussing the constraints and challenges in achieving this ideal.
- ### [[REVIEWS/Notes]]
- ### YELLOW CONCERNS
  background-color:: yellow
	- {{query (and [[ffd400]] [[RecSys_2024_paper_810]] )}}
	  collapsed:: true
- ### ❓️Questions
	- {{query (and [[question]] [[RecSys_2024_paper_810]] )[[question]]}}
	  query-table:: true
	  query-properties:: [:block]