An Improved Model for Depression Detection in Micro-Blog Social Network

Abstract

Social networks contain a tremendous amount of node and linkage data, providing unprecedented opportunities for a wide variety of fields. As the world's fourth largest disease, depression has become one of the most significant research subjects. Previously, a depression classifier has been proposed to classify the users in online social networks to be depressed or not, however, the classifier takes only node features into account and neglects the influence of linkages. This paper proposes an improved model to calculate the probability of a user being depressed, which is based on both node and linkage features. The linkage features are measured in two aspects: tie strength and interaction content analysis. Moreover, the propagation rule of depression is considered for improving the prediction accuracy. Finally, our experiments on the data derived from Sina Micro-blog shows that the highest accuracy of the improved model is 95%, increasing by 15% compared to the classifier with node features considered only. In this paper, it is well proved that adding linkage features analysis performs much better than node features analysis only. It also implies that tie strength and interaction content have different effects on depression probability estimation. Although this model is proposed for depression detection, the basic idea of linkage features analysis could be explicitly used in a wide scenario.