Approach Based on Bayesian Network and Ontology for Identifying Factors Impacting the States of People with Psychological Problems from Data on Social Media

Nowadays, social networks provide relevant information that is used in many contexts for different objectives. However, the major challenges remain at the level of processing this data, which is generated in a specific way. In this context, we propose in this paper a hybrid approach based on Bayesian network and ontology techniques for formalizing textual data published on social media by people with personality disorders. The objective of this task is to identify the main factors that have a significant impact on the state of sick persons. Our proposed approach is composed of three major steps: data collection and preprocessing, the construction of a set of Bayesian networks, and finally the incorporation of semantic components into the constructed networks. Our proposed approach takes advantage of both statistic and linguistic techniques, which can provide explainable and enriched results at multiple hierarchical levels. In addition, our approach addresses language issues like the evolution of the lexicon over time, the ellipsis phenomenon, etc. For the evaluation of our proposed approach, we have used two different methods, and in general, we achieved an accuracy rate equal to 83% for correct links prediction.