Mining Digital Traces of Facebook Activity for the Prediction of Individual Differences in Tendencies Toward Social Networks Use Disorder: A Machine Learning Approach.

Abstract

More than three billion users are currently on one of Meta’s online platforms with Facebook being still their most prominent social media service. It is well known that Facebook has designed a highly immersive social media service with the aim to prolong online time of its users, as this results in more digital footprints to be studied and monetized (via psychological targeting). In this context, it is debated if social media platforms can elicit addictive behaviors. In the present work, we demonstrate in N = 1,094 users that it is possible to predict from digital footprints of the Facebook users their self-reported addictive tendencies toward social media (R > 0.30) by applying machine-learning strategies. More specifically, we analyzed the predictive power of a set of models based on different sets of features extracted from digital traces, namely posting activity, language use, and page Likes. To maximize the predictive power of the models, we used an ensemble of linear and non-linear prediction algorithms. This work showed also sufficient accuracy rates (AUC above 0.70) in distinguishing between disordered and non-disordered social media users. In sum, individual differences in tendencies toward “social networks use disorder” can be inferred from digital traces left on the social media platform Facebook. Please note that the present work is limited by its cross-sectional design.