I should not recommend it to you even if you will like it: the ethics of recommender systems

Tang, Tiffany Y.; Winoto, Pinata

doi:10.1080/13614568.2015.1052099

Cited by 16 publications

(18 citation statements)

References 74 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…This raises an issue of content censorship with all its consequent implications. As noted in [14], what is algorithmically appropriate is not necessarily ethically appropriate. Should the RS be held liable for serving offensive or hazardous content?…”

Section: Towards Ethical Recommender Systemsmentioning

confidence: 99%

“…to a user's persona (gender, age, etc. ), as explained in [14]. Authors suggest that moral values are incorporated in system requirements, with users having control over the filtering process.…”

Section: Towards Ethical Recommender Systemsmentioning

confidence: 99%

“…With very few exceptions (e.g. [14]), contemporary RS do not employ any moral filtering of their output. This raises an issue of content censorship with all its consequent implications.…”

Section: Towards Ethical Recommender Systemsmentioning

confidence: 99%

“…The problems of ethics in recommender systems certainly go far beyond anonymization and relate to areas such as data collection and filtering, algorithmic opacity and biases, behaviour manipulation, A/B testing, etc. According to the recent study by Tang & Winoto [14], there exist only two publications (apart from [14]) that specifically address the problem of ethical recommendations. Therefore, this research direction deserves further attention.…”

Section: Introduction: Echoes From the Netflix Prizementioning

confidence: 99%

See 3 more Smart Citations

Recommender Systems from an Industrial and Ethical Perspective

Paraschakis

2016

Proceedings of the 10th ACM Conference on Recommender Systems

View full text Add to dashboard Cite

Over the recent years, a plethora of recommender systems (RS) have been proposed by academics. The degree of adoptability of these algorithms by industrial e-commerce platforms remains unclear. To get an insight into real-world recommendation engines, we survey more than 30 existing shopping cart solutions and compare the performance of popular recommendation algorithms on proprietary e-commerce datasets. Our results show that deployed systems rarely go beyond trivial "best seller" lists or very basic personalized recommendation algorithms, which nevertheless exhibit superior performance to more elaborate techniques both in our experiments and other related studies. We also perform chronological dataset splits to demonstrate the importance of preserving the sequence of events during evaluation, and the recency of events during training. The second part of our research is still ongoing and focuses on various ethical challenges that complicate the design of recommender systems. We believe that this direction of research remains mostly neglected despite its increasing impact on RS' quality and safety.

show abstract

Section: Towards Ethical Recommender Systemsmentioning

confidence: 99%

Section: Towards Ethical Recommender Systemsmentioning

confidence: 99%

“…With very few exceptions (e.g. [14]), contemporary RS do not employ any moral filtering of their output. This raises an issue of content censorship with all its consequent implications.…”

Section: Towards Ethical Recommender Systemsmentioning

confidence: 99%

Section: Introduction: Echoes From the Netflix Prizementioning

confidence: 99%

See 2 more Smart Citations

Recommender Systems from an Industrial and Ethical Perspective

Paraschakis

2016

Proceedings of the 10th ACM Conference on Recommender Systems

View full text Add to dashboard Cite

show abstract

“…According to the recent study by Tang & Winoto [193], there exist only three publications (namely, [169,186,193]) that specifically address ethics-aware recommendations. Still, these and similar works (e.g.…”

Section: From Algorithms To Ethicsmentioning

confidence: 99%

Algorithmic and Ethical Aspects of Recommender Systems in e-Commerce

Paraschakis¹

View full text Add to dashboard Cite

In memory of a beautiful soul ABSTRACTRecommender systems have become an integral part of virtually every e-commerce application on the web. The deployment of these expert systems has enabled users to quickly discover the products or services they need, at the same time increasing business revenues through better customer conversion. Remaining a very active research field since the mid-2000s, recommender systems have been modeled using a plethora of machine learning techniques. However, the adoptability of these models by industrial e-commerce platforms remains unclear.In this thesis, we assess the receptiveness of industrial platforms to algorithmic contributions of the research community by surveying more than 30 popular shopping cart solutions, and experimenting with various recommendation algorithms on proprietary e-commerce datasets.Another overlooked but important factor that complicates the design and use of recommender systems is their ethical implications. We provide a holistic view of these issues and summarize them in our ethical recommendation framework. This framework suggests new paradigm of ethics-awareness by design, and enables users to control sensitive moral aspects of recommendations via the proposed "ethical toolbox". The feasibility of this tool is supported by the results of our user study.Since the large part of moral implications stems from user profiling, we investigate algorithms capable of generating useragnostic recommendations based solely on a visited product page. We propose an ensemble learning scheme based on Thompson Sampling bandit policy, which models arms as base recommendation functions. We show how to adapt this algorithm to realistic situations when neither arm availability nor reward stationarity is guaranteed.

show abstract

Lack of the normative lens: Discussions on research on micro‐targeted ads explanation of Facebook

2019

Proceedings of the Association for Information Science and Tech

View full text Add to dashboard Cite

This paper observes that studies on the explanation of recommender systems lack a normative lens. Specifically, it uses Facebook's micro‐targeted advertisement explanations as a case for discussion, and shows how research mostly take a descriptive lens and focus on answering the what and how questions while the normative questions concerning how things should be are under explored. The paper argues that the normative and descriptive lenses are interdependent, and the lack of the normative lens contributes to inconsistent research suggestions. Explicitly putting on a normative lens is how researchers could avoid potential conflict among multiple normative goals.

show abstract

I should not recommend it to you even if you will like it: the ethics of recommender systems

Cited by 16 publications

References 74 publications

Recommender Systems from an Industrial and Ethical Perspective

Recommender Systems from an Industrial and Ethical Perspective

Algorithmic and Ethical Aspects of Recommender Systems in e-Commerce

Lack of the normative lens: Discussions on research on micro‐targeted ads explanation of Facebook

Contact Info

Product

Resources

About