Facebook versus Twitter: Differences in Self-Disclosure and Trait Prediction

Abstract

This study compares self-disclosure on Facebook and Twitter through the lens of demographic and psychological traits. Predictive evaluation reveals that language models trained on Facebook posts are more accurate at predicting age, gender, stress, and empathy than those trained on Twitter posts. Qualitative analyses of the underlying linguistic and demographic differences reveal that users are significantly more likely to disclose information about their family, personal concerns, and emotions and provide a more `honest' self-representation on Facebook. On the other hand, the same users significantly preferred to disclose their needs, drives, and ambitions on Twitter. The higher predictive performance of Facebook is also partly due to the greater volume of language on Facebook than Twitter -- Facebook and Twitter are equally good at predicting user traits when the same-sized language samples are used to train language models. We explore the implications of these differences in cross-platform user trait prediction.

What Twitter Profile and Posted Images Reveal about Depression and Anxiety

“…Compared to other psychological outcomes such as personality, these correlations are consistent e.g. (Segalin et al 2017;Reece and Danforth 2017;Jaidka et al 2018).…”

“…This is likely due to the fact that Twitter and Facebook have differences in language, both in terms of vocabulary (e.g. emoticons) and subject matter use (Jaidka et al 2018;Zhong et al 2017;Guntuku et al 2019). Importantly, these highly significant correlations (p < .01) nevertheless demonstrate that the Facebook prediction models encode significant mental health information that can be used to estimate the mental health status of Twitter users.…”

Self-Disclosure of Bullying Experiences and Social Support in Avatar Communication: Analysis of Verbal and Nonverbal Communications

“…emojis andemoticons (Kralj Novak et al 2015;Aldunate and González-Ibáñez 2017). People tend to disclose their emotion and sensitive information in closed cyber spaces with friends (not strangers) (Jaidka, Guntuku, and Ungar 2018). Consequently, bullied users may have disclosed their bullying experiences and very sensitive topics with their emotions.…”

Modeling and Visualizing Locus of Control with Facebook Language

“…The abundance of social media data presents researchers with a unique opportunity to profile users and communities from the language they write. Many researchers have explored users' social media language to infer user attributes including age and gender (Schwartz et al 2013;Jaidka, Guntuku, and Ungar 2018), personality (Plonsky, Erev, and others 2017;Rieman et al 2017), and mental and physical health (Jaidka, Guntuku, and Ungar 2018). However, it is not known whether the relationships between different user traits connect with each other, and whether these relationships can be inferred on the basis of language alone.…”