With its quality journalism the TX Group seeks to support a free and democratic society in which people can form their own opinions and are able to make informed choices. As Switzerland's strongest editorial network, we are aware of the power of speech which includes reproducing gender schemas. Such gender schemas can occur (even without any bad intent) if, for example, the majority of politicians who are interviewed for news articles are men even though that men and women are nearly equally represented in the parliament. In this talk I will mainly focus on a service which we kicked off a few years ago and got high attention in the newsrooms in the recent year: The Equality Alert. With this report our team provides transparency for journalists and the newsrooms in our company regarding gender-visibility in their media coverage. I will: (1) highlight the importance of such text analytics; (2) show how we have implemented the idea and delivered a now widely used report; and, (3) present how such a report can accelerate change by providing transparency.
Al Barsha Escorts $#$ O565212860 $#$ Escort Service In Al Barsha
[DSC Europe 22] Equality alert in newsrooms: Providing transparency about the visibility of gender in newspapers - Dominic Hercog
1. Equality alert in newsrooms:
Providing transparency about the visibility
of gender in newspapers
DSC Europe 2022
Dr. Dominic Herzog
Chief Data Scientist
TX Group AG / TX Services
dominic.herzog@tx.group
2. Why should we care about gender visibility
Draw (imaginize) a scientist…
3. Why should we care about gender visibility
DALL·E: Group of 5 female and male doctors with different skin color, van gogh style dark
gray background
4. Why should we care about gender visibility
girl's drawings of scientists depicted female scientists occur …
… more often in later decades
➢ women's representation in science has
risen substantially in the United States
➢ mass media increasingly depict female
scientists
Miller DI, Nolla KM, Eagly AH, Uttal DH. The Development of Children's Gender-Science Stereotypes: A Meta-analysis of 5
Decades of U.S. Draw-A-Scientist Studies. Child Dev. 2018 Nov;89(6):1943-1955. doi: 10.1111/cdev.13039. Epub 2018 Mar 20.
PMID: 29557555.
… less often among older girls.
➢ results may reflect that children
observe more male than female
scientists in their environments
5. Pla
TX Group and why we care about (gender) visibility
TX Group Portfolio
Tamedia paid media
20 Minuten commuter media
Goldbach advertising
Ventures marketplaces, job market, …
6. Why TX cares
Guidelines, principles and vision
Quality assurance and social
responsibility are two of the issues
we focus on to ensure we can always
live up to our journalistic
responsibility.
In 2021 - following an open letter
from 78 women journalists - a focus
was placed on diversity, inclusion
and equal opportunities. Including
gender balance in the workforce and
also in media coverage.
Business Reasons
Young females are
underrepresented in the (target)
audience while having
above-average shopping cart
values.
(internal) data shows that women
are more attracted to content that
thematizes or shows them.
Papers about gender bias in models
Google News 300-dimensional word2vec embedding
7. Why do we care
Readers
DALL·E: a park with 20 peoples reading newspapers in zurich or looking at their smartphone, comic style dark gray background
DALL·E: 10 journalists in a newsroom sitting at a round table with a TV screen which shows pie charts in the background, comic style on dark gray background
DALL·E: multiple newspaper articles and smartphones with news article on the screen, comic style on dark gray background
Newsrooms
Journalists
Articles
What we do:
❏ Quantitative insights
❏ Backup / neglect gut feelings
❏ Reveal (unconscious) biases in our product
❏ Reality check and providing new tools to align
product with guidelines, visions & business
opportunities
What is a no-go:
❏ Qualitative judgement of a single article
=> journalistic freedom within TX guidelines
❏ Change / “Debias” content
8. Visibility I: Report
Equality Alert Daily analysis and report on women's visibility in Tamedia publications
❏ Itʼs used and monitored
in the newsrooms
(among other aspects)
❏ Potential to optimize
how we report in order to
integrate it better into
daily work
9. Defining “Gender” for this presentation
Gender visibility in articles
➢ Identify names in articles
➢ Guess gender (female | male)
based on the personʼs first
name
Gender of the reader
➢ Hard facts from registration
➢ If not available then
predicting gender based on
reading behavior (using
reading behavior of users
with hard facts)
Gender of the author
➢ Guess gender (female | male)
based on the authorʼs first
name
! Using binary female-male annotation and analyses is based on availability of the data.
! This approach is not neglecting of the gender ʻdiversʼ nor should it imply that all persons in the binary
groups female-male are acting the same.
10. Visibility I: Report
Equality Alert Daily analysis & report of women's visibility in Tamedia publications (& 20min)
❏ The visibility is quite noisy on a daily basis and fluctuates between 15% and 30% in 2021-2022
Visibility
of
women
in
articles
50%
10%
30%
40%
20%
11. Visibility II: Visibility of women in our publications | Over time
❏ Visibility of women in articles has increased from 17% to 24% since 2014; no additional effect since 2021
❏ Visibility of women in articles written internally [black] coincides with the
Visibility of women in articles written by the agency SDA [green].
Visibility: Measured by the people named gender_author & share_female: Binary gender mapping based on first name
gender_author: Gender of the author of the article (for articles written internally) or name of the news agency
Visibility
of
women
in
articles
30%
10%
20%
25%
15%
Do our initiatives show the desired impact and trend?
12. Visibility III: Visibility of women in our publications | Over time & split by gender of the autor
❏ Persistent difference in the breakdown by gender of the author
❏ If an article was written by a woman in 2022, 33% of the names mentioned in the text are female.
If an article was written by a man in 2022, 24% of the names mentioned in the text are female.
Visibility: Measured by the people named gender_author & share_female: Binary gender mapping based on first name
gender_author: Gender of the author of the article (for articles written internally) or name of the news agency
Does the gender of the author matters regarding the visibility of women?
Visibility
of
women
in
articles
35%
15%
25%
30%
20%
13. Visibility III: Visibility of women in our publications | Over time & split by gender of the autor
Yes, but… this might depend on the section…
Section: Switzerland
Section: Sonntagszeitung
14. Visibility III: Visibility of women in our publications | Over time & split by gender of the autor
Yes, but… I mean the other sections…
Section: Foreign Affairs
Section: Culture
15. Visibility III: Visibility of women in our publications | Over time & split by gender of the autor
Yes, but… I mean the other sections…
Section: Sports
Section: Sports - Soccer
16. Visibility III: Visibility of women in our publications | Over time & split by gender of the autor
Yes, but… this is news-paper dependent…
20min: German-Speaking Region
20min: French-Speaking Region
17. Visibility III: Visibility of women in our publications | Over time & split by gender of the autor
❏ The insights are persistent over sections as well as across newspapers
❏ The only exception is the section about foreign affairs
18. Visibility IV: Testing the hypothesis that articles about women are more attractive to female readers
❏ 24.9% of women…
❏ 20.9% of men…
... read articles which mainly name women.
❏ Articles that mainly mention women [men] reach a
female readership share of 43.3% [37.8%]
80%
0%
40%
60%
20%
80%
0%
40%
60%
20%
100%
Percentage
of
articles
read
100%
gender_reader: Gender of the reader
dominant_gender_article: Dominant gender in the article, measured by the persons named
19. Visibility V: Text Corpus analysis & (fe)male annotated words
A ➡ B is like Y ➡ Z: Examples from our own text corpus
Man (Mann) ➡ King (König) is like Woman (Frau) ➡ Queen (Königin)
Man (Mann) ➡ Programmer (Programmierer) is like Woman (Frau) ➡ Influence (Influencerin)
Man (Mann) ➡ Farmer (Bauer) is like Woman (Frau) ➡ Farmer (Bäuerin)
Man (Mann) ➡ Doctor (Arzt) is like Woman (Frau) ➡ Doctor (Ärztin)
Word2Vec based on 342ʼ500 internally written articles, published in tagesanzeiger (print/online) since 2012
20. Visibility V: Text Corpus analysis & (fe)male annotated words
Male - Female annotated words
Word2Vec based on 342ʼ500 internally written articles, published in tagesanzeiger (print/online) since 2012
gender-vector: [('Frau', 'Mann'), ('Tochter', 'Sohn'), ('Mädchen', 'Junge')]
Dominic’s
annotation
guess based
on his “socialization-learned”
Stereotypes:
Female annotated
Male annotated
woman
handbag
lady
hair
wipe
silly
clean
slight
repair
handcraft
educated
household
ambitious
backpack
gentleman
workplace
man
21. Wrap up
➢ Qualitative: Journalists are the experts on identifying topics and writing the article
➢
➢ Quantitative: Data Scientists are experts to analyze the text corpus and articles
➢ Visibility of women increased in the past 10 years from 17% to 24% but did not further increase since
2021
➢ There is a substantial and persistent gap in the visibility provided between female and male authors.
Overall but also across newspapers and (sub-)sections female journalists provide more visibility to
females.
➢ Gender stereotypes which are present in society are to some extent reproduced in our own text
corpus.
➢ This is work in progress & partnering with newsroom is key
➢
➢ The Equality Alert focused so far on gender. However, the approaches shown can be applied to other
dimensions (of potential discrimination) such as origin, age, skin color or income
22. Change through data
Data and Data Science tools can reveal hidden patterns and make biases visible
Visibility leads to awareness
Awareness invites to reflect the status quo based on our values & vision
Our values & vision can guide us towards needed change
Change through data - to the good of all of us!
dominic.herzog@tx.group