CoChat

Bias in Search and Recommender Systems

Baeza-Yates, Ricardo · 2020 · Fourteenth ACM Conference on Recommender Systems
Cited by 76

The DOI matches its Crossref record; 2 of the 2 references it cites were found in a registry or an index.

Abstract

We explore the vicious cycle of bias on the Web related to search and recommender systems. The first bias is activity bias [1], called by Nielsen participation inequality in Internet [3]. This means that when sampling content, data will have many different types of bias such as gender, language, topic, etc. During the design and implementation of the system, biases might be added, and we call those the true algorithmic bias. In addition, when evaluating the system we may have been biased and hence that may also be reflected in algorithmic bias. The design of the interaction also matters, adding several other biases. All of them affect the data that is gathered for optimizing and personalizing the system [1]. Finally, our own cognitive biases also taint the interaction data, including confirmation bias and other behavioral biases. In all these cases we may need to debias the input data, to use techniques that can handle biased data (e.g., machine learning algorithms tailored for that), or debias the output (when we have already lost information). Most systems are optimized by using implicit user feedback, that is, clicks or other trackable user interactions. However, those interactions are biased to the choices that such systems offer to their users, as clicks can only be done on things that are shown to us. Hence, these feedback loops are tainted by presentation or exposure bias [1]. The most well-known problem related to this bias is called the filter bubble [4] or the echo chamber effect. Solutions to this problem include the explore and exploit paradigm (e.g., learning the world), diversity, novelty, and serendipity. Depending on the exploration technique, the bubble is smaller or bigger [2]. On the other hand, too much exploration also reduces short-term revenue and hence is usually bounded. However, we believe that recommender systems could improve their long-term revenue if significantly more exploration is performed, probably diminishing at the same time the tension between user experience and monetization. This is good for the recommender system but also creates more fair and healthy digital markets for everyone, users and publishers/sellers.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1145/3383313.3418435

Cites

2 of 2 references verified0 citation errors · 0 not checkable

Statistics

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Cited by

See all 76

Metadata

Authors
Baeza-Yates, R.
Venue
Fourteenth ACM Conference on Recommender Systems
Publisher
ACM
Published
2020-09-22
Pages
2-2
OpenAlex
W3088924330
Cited by
76
References
2
FWCI
9.16
Type
conference paper
Language
en
Open access
closed

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.