Predicting delay discounting from heterogeneous social media data

Abstract

Delay discounting, a behavioral measure of impulsivity, is often used to quantify the human tendency to choose a smaller, sooner reward (e.g., $1 today) over a larger, later reward ($2 tomorrow). Delay discounting and its relation to human decision-making is a hot topic in economics and behavior science since pitting the demands of long-term goals against short-term desires is among the most difficult tasks in human decision-making (Hirsh et al. in J Res Personal 42:1646, 2008). Previously, questionnaire-based surveys were used to assess someone’s delay discounting rate (DDR). In this research, we develop a computational model to automatically predict DDR from diverse types of social media data such as likes and posts. We explore different unsupervised feature learning methods to take advantage of a large amount of unsupervised social media data. Based on our evaluation, the best model learned from social media likes achieved 0.634 ROC AUC, which 18% better than the baseline model that did not use unsupervised feature learning. The best model learned from social media posts achieved 0.641 ROC AUC, which is 24% better than the baseline that did not use unsupervised feature learning. We also employed unsupervised feature fusion to combine heterogeneous user data such as likes and posts together to further improve system performance. The final combined model outperformed the baseline that did not use unsupervised feature learning by 30%. Finally, we conducted additional analysis to uncover interesting correlation patterns between a person’s social media behavior and his/her DDR.