The Effect of Population and "Structural" Biases on Social Media-based Algorithms - A Case Study ...
summary
Compares open-source text-based model and network-based model on geo-tagged Twitter across the urban–rural spectrum, using 100 km error as the accuracy threshold. Finds urban precision more than twice rural for the text model, and rebalancing to match US population barely improves rural text performance even when training only on rural tweets; network performance improves dramatically with rebalancing and can exceed urban precision, so structural algorithm design—not just data skew—drives some bias.
AI-written summary · about
chapters (8)
- 0:00Introduction and motivation
- 1:51Urban-rural bias in social media
- 3:16Output bias in geolocation algorithms
- 6:28Algorithm types and data methods
- 9:19Results for research questions
- 13:11Key takeaways and structural bias
- 16:46Hybrid methods and conclusion
- 17:06Q&A session
related talks
50mWhat It Takes to Control Societal Bias in Natural Language Processing
38mSocial Implications of Bias in Machine Learning
38mSocial Implications of Bias in Machine Learning
30mDisambiguating Explicit and Implicit Geographic References...
22mMaking Natural Language Processing Robust to Sociolinguistic Variation
48mChallenges For Causal Inference On Digital Platforms
56mGlobal Bias in AI: When Western Data Shapes the World
53mIt’s Not Fair! Detecting Algorithmic Bias with Open Source Tools