Pia Pachinger

The size of AustroTox

2024 ACL  Findings 
AustroTox: A Dataset for Target-Based Austrian German Offensive Language Detection 
Pia Pachinger, Janis Goldzycher, Anna Maria Planitzer, Wojciech Kusa, Allan Hanbury, Julia Neidhardt

We create the first dataset with a focus on Austrian German toxic language online and the first German dataset on toxicity featuring annotated spans facilitating explainability and error analysis. We find that LLMs not fine-tuned on AustroTox fail to recognize country-specific vulgar language and the targets of toxic statements (as of 1 / 2024). 
 

2025 EMNLP NLPerspectives
A Disaggregated Dataset on English Offensiveness Containing Spans
Pia Pachinger, Janis Goldzycher, Anna Maria Planitzer, Julia Neidhardt, Allan Hanbury

We re-annotate a subset of posts of the Jigsaw Toxic Comment Classification Challenge and provide disaggregated toxicity labels and spans. We find that five annotations per instance allow for fine distinctions between genuine disagreement and that arising from annotation error or inconsistency. Disagreement is especially high in cases of toxic statements involving non-human targets.

Disagreement in toxicity labels

Annotation framework

Disagreement in marking vulgar spans

Disagreement in marking targets of offensive statements

Classes in the aggregated dataset

Comparison to Jigsaw annotations

2023 EACL C3NLP 
Toward Disambiguating the Definitions of Abusive, Offensive, Toxic, and Uncivil Comments
Pia Pachinger, Anna Maria Planitzer, Julia Neidhardt, Allan Hanbury

We find that researchers studying harmful language online treat abusiveness, offensiveness, and toxicity interchangeably as sub-concept of one another. While social science literature frequently employs incivility with similar meaning, computer science research underutilizes this term, suggesting both fields would benefit from terminological unification. We compile and analyse the distinct definitions researchers use for these concepts to facilitate future unification efforts across and within disciplines.

Counts of terms used to describe toxicity, offensiveness and incivility i​n different papers on content moderation across disciplines

Co-authors and two-step-coauthors of an expert

2022 TU Wien
A Recommender System for Scientific Referees Based on Bibliographic Databases and Knowledge Graphs
Pia Pachinger, Georg Gottlob, Joël Ouaknine, Glenn Starkman, Matt Rainey, Emanuel Sallinger

We implement multiple recommender systems for scientific expert search using two-step coauthorship and publication venue overlap. Qualitative evaluation by 15 established computer science researchers showed the best system recommended a mean of 5 excellent and 2.8 suitable experts per 10 recommendations.

Workshop on Online Abuse and Harms (ACL) 2025
Alignment by Disagreement? Investigating Individual and Sociodemographic Variability in the Perception of Harmful Language 
Pia Pachinger, Anna Maria Planitzer, Allan Hanbury, Julia Neidhardt, Sophie Lecheler

International Communications Association Conference 2025
Like Walking a Tightrope: User-Centric Perspectives on Automated Content Moderation
Anna Maria Planitzer, Sophie Lecheler, Svenja Schäfer, Pia Pachinger (presented by Anna)

COMPTEXT 2025 
Incorporating User Perceptions of Online Norm Violations in Toxicity Detection Models
Pia Pachinger, Anna Maria Planitzer, Allan Hanbury, Julia Neidhardt, Rebekah Wegener, Sophie Lecheler

2024 NAACL, Student Research Workshop 
Best PhD proposal 
User-Centric Offensive Text Detection in Culture Specific Contexts: 
A PhD Proposal 
Pia Pachinger

2022 Inria, Paris
Design + interaction + AI hackathon 
Second Prize for That's life. 
Artifact publicly exhibited at Le Bis in Paris
Anaïs Cambou*, Anthonin Gourichon*, Fengyu Li*, Xiaoning Meng*, Pia Pachinger* (* equal contribution)

That's Life.

TU Wien ​2022 - 2026 
Faculty of Informatics
Prae-Doc researcher
TACo:  User-centric content moderation 
BrAIn: Domain-specific information extraction 
VHH: Evaluation of machine translation models 

Austrian Institute of Technology 2019 - 2020
Center for Cyber Security 
Researcher
Pre-training and evaluation of CNNs and LSTMs for anomaly detection in time series of system log data

TU Wien  2021 - 2022 
Faculty of Informatics
Student researcher
Implementation of recommender system for scientific referees

University of Vienna 2018-2020
Faculty of Mathematics
Teaching Assistant