Finding Potential Empathizers in an Online Mental Health Community: A Deep Graph Embedding Approach

Empathic support in online health communities (OHCs) can significantly help patients to feel less suffering and pain, fight against their diseases, and eventually recover. However, effective empathic communications in real-world OHCs are of low percentage. In this paper, we use recommendation techniques to match support-seeking posts with potential empathizers in a real-world online mental health community. Inspired by complex network analysis, we construct information-rich heterogenous network to model OHC. A graph embedding learning algorithm is devised to learn effective user profile embeddings from various information sources in heterogenous network, including node content feature, local neighborhood and sub-community structures. Compared with baseline methods, our approach achieves the best-performing recommendation results and fast computing speed.