Private traits and attributes are predictable from digital records of human behavior

By Michal Kosinski, David J. Stillwell, Thore Graepel. 2013

2,640 citations recorded by the source corpus.

We show that easily accessible digital records of behavior, Facebook Likes, can be used to automatically and accurately predict a range of highly sensitive personal attributes including: sexual orientation, ethnicity, religious and political views, personality traits, intelligence, happiness, use of addictive substances, parental separation, age, and gender. The analysis presented is based on a dataset of over 58,000 volunteers who provided their Facebook Likes, detailed demographic profiles, and the results of several psychometric tests. The proposed model uses dimensionality reduction for preprocessing the Likes data, which are then entered into logistic/linear regression to predict individual psychodemographic profiles from Likes. The model correctly discriminates between homosexual and heterosexual men in 88% of cases, African Americans and Caucasian Americans in 95% of cases, and between Democrat and Republican in 85% of cases. For the personality trait “Openness,” prediction accuracy is close to the test–retest accuracy of a standard personality test. We give examples of associations between attributes and Likes and discuss implications for online personalization and privacy.

Subfields: Clinical Psychology, Social Psychology, Sociology and Political Science, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Clinical Psychology

In Social Psychology

In Sociology and Political Science

In Safety, Robustness & Interpretability

Citations in the corpus

Ethical and social risks of harm from Language Models (2021, background): “Leveraging language processing tools and large public datasets to infer private traits is an active area of research (Kosinski et al., 2013; Park et al., 2015; Quercia et al., 2011; Youyou et al., 2015).”
On formalizing fairness in prediction with machine learning (2017, background): “[44] show that it is possible to accurately predict a range of highly sensitive personal aributes including: sexual orientation, ethnicity, religious views, age, and gender from easily accessible digital records of behaviour.”
Explaining Classification Models Built on High-Dimensional Sparse Data (2016, background): “What you “like” on Facebook for example allows the prediction of personal characteristics remarkably well (Kosinski et al., 2013), as well as predicting product interest or credit default behavior (De Cnudde et al., 2015).”
Taxonomy of Risks posed by Language Models (2022, background): “Leveraging language processing tools and large public datasets to infer protected and other personal traits is an active area of research [107, 140, 146, 200], despite serious ethical concerns [6, 185].”
Do social explanations work?: studying and modeling the effects of social explanations in recommender systems (2013, background): “While we discussed them mainly from the perspective of utility, we have to be mindful of any social norms or privacy expectations these explanations might violate, or personal information they might disclose [13].”
A Brief History of Artificial Intelligence: On the Past, Present, and Future of Artificial Intelligence (2019, neutral): “What such rules can, however, not avoid is the deliberate hacking of AI systems, the unwanted use of such systems for micro-targeting based on personality traits, 16 or the generation of fake news.”
Predicting the Big 5 personality traits from digital footprints on social media: A meta-analysis (2018, methodology): “Among studies using MyPersonality data collected on Facebook, we included in the analyses those which examined the most comprehensive set of digital footprints, and in case they examined the same set of digital footprints, we selected those with the largest sample (Farnadi et al., 2016 Study 1; Kosi”
Tweeting From Left to Right (2015, background): “Although ideology is one of the key predictors of social and political behavior ( Jost, 2006), only a handful of studies have attempted to measure ideology using social-media data (e.g., Back et al., 2010; Conover et al., 2011; Kosinski et al., 2013).”
Attribute Inference Attacks in Online Social Networks (2018, background): “Our results indicate that our graph-based attack is a better way to leverage behavior structures, compared to LG-B-I and LG-B-II, which flatten the behavior structures into feature vectors.”