Adimen Artifiziala Joko Patologikoaren Arriskua Aurresateko

Abstract

In this work, we propose artificial intelligence techniques based on deep neural learning to enable the early detection of gambling addiction risk indicators in social media. To achieve this, we analyze sequences of user posts to identify latent signals within the text. One of the main challenges is the class imbalance in the data, as the group of users with gambling addiction is 13 times smaller than the control group. To address this issue, we have developed a sample selection strategy aimed at balancing Precision and Recall. Additionally, we designed a system capable of modeling the evolution of risk indicators over time, with the goal of detecting them using as few messages as possible. Our contributions are framed within the field of ‘Language Analysis and Processing,’ with applications in the medical domain.