Loading...
Gender bias in machine learning for sentiment analysis
Thelwall, Mike
Thelwall, Mike
Authors
Editors
Other contributors
Affiliation
Epub Date
Issue Date
2018-01-01
Submitted date
Alternative
Abstract
Purpose: This paper investigates whether machine learning induces gender biases in the sense of results that are more accurate for male authors than for female authors. It also investigates whether training separate male and female variants could improve the accuracy of machine learning for sentiment analysis. Design/methodology/approach: This article uses ratings-balanced sets of reviews of restaurants and hotels (3 sets) to train algorithms with and without gender selection. Findings: Accuracy is higher on female-authored reviews than on male-authored reviews for all data sets, so applications of sentiment analysis using mixed gender datasets will over represent the opinions of women. Training on same gender data improves performance less than having additional data from both genders. Practical implications: End users of sentiment analysis should be aware that its small gender biases can affect the conclusions drawn from it and apply correction factors when necessary. Users of systems that incorporate sentiment analysis should be aware that performance will vary by author gender. Developers do not need to create gender-specific algorithms unless they have more training data than their system can cope with. Originality/value: This is the first demonstration of gender bias in machine learning sentiment analysis.
Citation
Thelwall, M. (2018), "Gender bias in machine learning for sentiment analysis", Online Information Review, Vol. 42 No. 3, pp. 343-354. https://doi.org/10.1108/OIR-05-2017-0153
Publisher
Research Unit
PubMed ID
PubMed Central ID
Embedded videos
Type
Journal article
Language
en
Description
This is an accepted manuscript of an article published by Emerald Publishing Limited in Online Information Review on 01/01/2018, available online: https://doi.org/10.1108/OIR-05-2017-0153
The accepted version of the publication may differ from the final published version.
Series/Report no.
ISSN
1468-4527