Digitize! – Computational Social Science in der digitalen und sozialen Transformation

Sylvia Kritzinger, Katharina Pfaff, Julia Barta, Jana Bernhard, Hajo Boomgaarden, Anja Eder, Nikolaus Forgó, Filip Paspalj, Claudia Plant, Barbara Prainsack, Dimitri Prandner, Simon Rittel, Martin Teuffenbach, Sebastian TschiatschekZeitschrift für Hochschulentwicklung 2023 · OpenReview

Abstract

Die Digitalisierung bringt neben vielen Vorteilen für die sozialwissenschaftliche Forschung und die Lehre auch neue Herausforderungen mit sich. Die Weiterentwicklung digitaler Datenerhebungs- und Analyseverfahren zur Analyse und Gestaltung gesellschaftlicher und politischer Transformationsprozesse muss im Fokus der Zukunftsausrichtung sozialwissenschaftlicher Hochschullehre und Forschung stehen. Neue Datenformate und Praktiken erfordern neue forschungsethische und datenschutzrechtliche Praktiken und Standards. Im Zentrum von Computational Social Science steht die juristisch und ethisch fundierte, reflektierte Nutzbarmachung von digitalen Forschungsdaten und Analyseverfahren und ihre Vermittlung.

Introduction

Social science is currently navigating a transition between traditional rigorous methodology and the expansive "Big Data" era. While digitalization offers unprecedented access to information, it introduces significant hurdles: traditional survey response rates are declining, and "found" digital data (such as social media) often lacks representativeness and ethical oversight. Furthermore, most advanced Natural Language Processing (NLP) tools are optimized for English and commercial use, leaving a gap for researchers working with regional languages or public-interest questions. The Digitize! project addresses these challenges by building a representative, ethically sound, and technologically advanced infrastructure for social research in the 21st century.

Methods

The project employs an interdisciplinary framework across four core pillars:

  • Social Science: Implementing "push-to-web" designs, where participants are recruited offline via population registers but respond online, maintaining representativeness while reducing costs.
  • Data Science: Developing scalable algorithms for "hybrid" data types (e.g., ordinal survey scales) and using generative neural networks to impute missing survey data. This includes dynamic surveys that adapt based on previous answers to reduce respondent fatigue.
  • NLP & Text Analysis: Creating the METEOR database to standardize German text analysis tools and training localized language models specifically on Austrian news corpora.
  • Ethics & Law: Proposing a "Data Solidarity" framework that categorizes data use based on its "public value" rather than just the data type (personal vs. non-personal).

Datasets

The project utilized and generated several key resources to validate its framework:

  • Digitize! Online Panel Pilot: A representative Austrian panel involving multiple survey waves to track social trends.
  • Austrian Social Survey (SSÖ) & ISSP: Modernized versions of these longitudinal studies transitioned to a digital-first format.
  • Austrian News Corpus: Millions of news articles used to train a transparent, localized language model.
  • METEOR Database: A curated collection of 44 automated text analysis tools (ATAM) for German-language research.
Respondent participation and recruitment across the five waves of the Digitize! Online Panel Pilot.

Key Insights and Ideas

  • The Incentive Trade-off: Research into recruitment found that "unconditional" vouchers (given upfront) maximize immediate response rates, whereas "conditional" vouchers (promised after completion) are more effective at securing long-term panel members.
  • Data Solidarity: The authors argue for a regulatory matrix where high-public-value, low-risk research receives legal "fast-tracks," distinguishing public-interest science from high-risk commercial data mining.
  • The "Austrian" Language Gap: Standard Large Language Models (LLMs) often miss regional nuances. By training smaller, transparent models on local news, the team ensures scientific reproducibility and avoids corporate dependency.
  • The Teaching Lag: A survey of methodology teachers revealed that Computational Social Science (CSS) accounts for only 5.2% of current teaching, highlighting a critical gap between research needs and university curricula.

Key Results

  • Legal Impact: The project influenced Austrian legislation regarding EU Directive 2019/790, leading to broader Text and Data Mining (TDM) privileges for researchers in the Austrian Copyright Act (§ 42h UrhG).
  • Ethical Tooling: The team launched an online tool (pluto.univie.ac.at) that uses 21 weighted questions to calculate the "public value" of a data project, assisting researchers with ethical compliance.
  • Technical Innovation: Development of a hierarchical topic generation algorithm allowed for more nuanced automated categorization of open-ended survey responses compared to standard flat models.

Limitations and Future Work

While the localized language models offer transparency, they remain smaller than those produced by "Big Tech." Future work will focus on deeper integration of LLMs into social science workflows, specifically for designing "human-like" adaptive surveys. Additionally, the project aims to bridge the "teaching gap" by expanding Open Educational Resources (OERs) to ensure the next generation of social scientists is code-literate.

Summary updated

Abstract

Die Digitalisierung bringt neben vielen Vorteilen für die sozialwissenschaftliche Forschung und die Lehre auch neue Herausforderungen mit sich. Die Weiterentwicklung digitaler Datenerhebungs- und Analyseverfahren zur Analyse und Gestaltung gesellschaftlicher und politischer Transformationsprozesse muss im Fokus der Zukunftsausrichtung sozialwissenschaftlicher Hochschullehre und Forschung stehen. Neue Datenformate und Praktiken erfordern neue forschungsethische und datenschutzrechtliche Praktiken und Standards. Im Zentrum von Computational Social Science steht die juristisch und ethisch fundierte, reflektierte Nutzbarmachung von digitalen Forschungsdaten und Analyseverfahren und ihre Vermittlung.
BibTeX
@inproceedings{kritzinger2023digitize,
  author = {Sylvia Kritzinger and Katharina Pfaff and Julia Barta and Jana Bernhard and Hajo Boomgaarden and Anja Eder and Nikolaus Forgó and Filip Paspalj and Claudia Plant and Barbara Prainsack and Dimitri Prandner and Simon Rittel and Martin Teuffenbach and Sebastian Tschiatschek},
  title = {{Digitize! – Computational Social Science in der digitalen und sozialen Transformation}},
  booktitle = {Zeitschrift für Hochschulentwicklung},
  year = {2023},
  url = {https://openreview.net/forum?id=lwFTUfjb7S},
}
Figure