FairSearch: A Tool For Fairness in Ranked Search Results

Zehlike, Meike; Sühr, Tom; Castillo, Carlos; Kitanovski, Ivan

doi:10.1145/3366424.3383534

Cited by 14 publications

(6 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We experiment with a state-of-the-art algorithm called Fair Adversarial DirectRanker (AdvDR in the rest of this paper), which showed good results on commonly used fairness datasets [4], a Debiasing Classifier (AdvCls) based on gradient reversal [17,5,45,31], a fair listwise ranker (DELTR [50]) and the Variational Fair Autoencoder [28] 2 (VFAE). All these methods were also constrained for explicit computation of correction vectors.…”

Section: Methodsmentioning

confidence: 99%

“…We show this by focusing our experimental validation on extending four different fair representation learning approaches so that they may compute correction vectors. Our experimentation on a state-of-the-art fair ranking model (AdvDR [4]), a fair classifier (AdvCls [45,5,31]), the aforementioned Variational Fair Autoencoder (VFAE [28]) and a listwise ranker (DELTR [50]).…”

Section: Explicit Computation Of Correction Vectors With Feedforward ...mentioning

confidence: 99%

“…whether it is relying on sensitive, law-protected information such as ethnicity or gender to undertake its decisions. The usual approach here is to either constrain the output of the network [50,36] so that it is fair under some definition or to remove information about the sensitive attribute from the model's internal representations [45,4,31,28]. The latter techniques might be referred to as "fair representation learning" [51].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Fair Interpretable Representation Learning with Correction Vectors

Cerrato¹,

Coronel²,

Köppel³

et al. 2022

Preprint

View full text Add to dashboard Cite

Neural network architectures have been extensively employed in the fair representation learning setting, where the objective is to learn a new representation for a given vector which is independent of sensitive information. Various representation debiasing techniques have been proposed in the literature. However, as neural networks are inherently opaque, these methods are hard to comprehend, which limits their usefulness. We propose a new framework for fair representation learning that is centered around the learning of "correction vectors", which have the same dimensionality as the given data vectors. Correction vectors may be computed either explicitly via architectural constraints or implicitly by training an invertible model based on Normalizing Flows. We show experimentally that several fair representation learning models constrained in such a way do not exhibit losses in ranking or classification performance. Furthermore, we demonstrate that state-of-the-art results can be achieved by the invertible model. Finally, we discuss the law standing of our methodology in light of recent legislation in the European Union.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Explicit Computation Of Correction Vectors With Feedforward ...mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Fair Interpretable Representation Learning with Correction Vectors

Cerrato¹,

Coronel²,

Köppel³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…We implemented the presented correction and algorithms into FairSearch, an open-source API for fairness in ranked search results [3]. The code can be found at https://github.com/fair-search.…”

Section: Methodsmentioning

confidence: 99%

A Note on the Significance Adjustment for FA*IR with Two Protected Groups

Zehlike,

Sühr,

Castillo

2020

Preprint

Self Cite

View full text Add to dashboard Cite

In this report we provide an improvement of the significance adjustment from the FA*IR algorithm in Zehlike et al. [2], which did not work for very short rankings in combination with a low minimum proportion 𝑝 for the protected group. We show how the minimum number of protected candidates per ranking position can be calculated exactly and provide a mapping from the continuous space of significance levels (𝛼) to a discrete space of tables, which allows us to find 𝛼 corr using a binary search heuristic.

show abstract

“…FairSearch [93] is an open source API to provide fair search results, which is designed as stand-alone libraries (supported in Python and Java) and plugins of Elasticsearch (supported in Java). Users can run FairSearch together with their own datasets once these have been formatted as required.…”

Section: Fair Searchmentioning

confidence: 99%

Fairness in Ranking: A Survey

Zehlike,

Yang,

Stoyanovich

2021

Preprint

Self Cite

View full text Add to dashboard Cite

In the past few years, there has been much work on incorporating fairness requirements into algorithmic rankers, with contributions coming from the data management, algorithms, information retrieval, and recommender systems communities. In this survey we give a systematic overview of this work, offering a broad perspective that connects formalizations and algorithmic approaches across subfields.An important contribution of our work is in developing a common narrative around the value frameworks that motivate specific fairness-enhancing interventions in ranking. This allows us to unify the presentation of mitigation objectives and of algorithmic techniques to help meet those objectives or identify trade-offs.

show abstract

FairSearch: A Tool For Fairness in Ranked Search Results

Cited by 14 publications

References 10 publications

Fair Interpretable Representation Learning with Correction Vectors

Fair Interpretable Representation Learning with Correction Vectors

A Note on the Significance Adjustment for FA*IR with Two Protected Groups

Fairness in Ranking: A Survey

Contact Info

Product

Resources

About