2 min

1st Place for Detecting Sexism in Social Networks

St. Pölten UAS and AIT Develop Methods for Detecting Sexist Content Online and Achieve Excellent Results in EXIST 2023 Benchmark

1st Place for Detecting Sexism in Social Networks
Copyright: https://www.pexels.com/

Researchers from the IC\M/T (St. Pölten University of Applied Sciences) and the AIT Austrian Institute of Technology developed methods within the international EXIST Benchmarks (sEXism Identification in Social neTworks) that enable the automatic detection of sexist content in tweets, thereby achieving excellent results in multiple categories. Congratulations!

Detection & Classification of Sexist Content

Sexism is an omnipresent societal issue that has escalated in social media in particular. Recent studies show that especially women in the public domain face significant challenges. Social media act as an accelerator that lowers the inhibition threshold for verbal harassment.

The EXIST 2023 Benchmark focused on the detection and classification of sexist content in social media texts in English and Spanish. The team already successfully participated in the EXIST 2021 Benchmark and achieved the third place in the competition. "The tasks this year were far more complex than in 2021," says Djordje Slijepčević, researcher in the Media Computing research group at the IC\M/T. "Therefore, we are delighted to have achieved the 1st, 2nd, and 3rd places in two out of three tasks."

The EXIST 2023 Tasks

  • Task 1: Binary detection of sexism: Systems must determine whether a specific tweet contains sexist expressions or behaviours.
  • Task 2: Ternary classification of sexist content based on the authors’ intent.
    St. Pölten UAS/AIT ranking: 1st place for Spanish content & 2nd place for Spanish and English content.
  • Task 3: Multi-label classification of sexist content into one or more of the following categories: ideology and inequality, stereotyping and dominance, objectification, sexual violence, and misogyny and non-sexual violence.
    St. Pölten UAS/AIT ranking: 3rd place for Spanish and English content.

Read detailed descriptions of the individual tasks here.

Our Team at the EXIST 2023

The team includes the following colleagues from the St. Pölten University of Applied Sciences and the AIT (in alphabetical order):

  • Andreas Babic (FHSTP)
  • Jaqu Böck (FHSTP)
  • Daria Liakhovets (AIT)
  • Nathanya Queby Satriani (FHSTP)
  • Alexander Schindler (AIT)
  • Mina Schütz (AIT)
  • Djordje Slijepčević (FHSTP)
  • Matthias Zeppelzauer (FHSTP)

The CLEF 2023 (Conference and Labs of the Evaluation Forum), where the individual contributions will be presented, takes place in September in Thessaloniki, Greece.

Taking a Clear Stance against Hate Postings

"The methods we developed for the EXIST 2023 originated from the WWTF-funded project Counter Speech – Young People Against Online Hate and are used in this project in modified form (for German content)," explains Djordje Slijepčević.

The project focuses on the development of computer-assisted strategies to support counter speech in social media and contributes to combating hate online. Conventional measures such as removing comments, excluding individuals from social networks, or flagging hate posts do not achieve the desired results, and committed users often feel powerless or unsure how to respond appropriately to inappropriate statements without potentially becoming victims themselves. This is where the "Counter Speech" project comes in: To encourage young people to perform counter speech, the project team relies on automated methods for data analysis and generation. The idea is to increase the visibility of committed young users who stand up against hate comments and to make the effectiveness of their messages measurable.

You want to know more? Feel free to ask!
FH-Prof. Priv.-Doz. Dipl.-Ing. Mag. Dr. Zeppelzauer Matthias

FH-Prof. Priv.-Doz. Dipl.-Ing. Mag. Dr. Matthias Zeppelzauer

Head of
Media Computing Research Group
Institute of Creative\Media/Technologies
Department of Media and Digital Technologies