Health Status Detection via Temporal-Spatial Factor Graph Model in Medical Social Networks

Applications of Social Networks is very popular in various fields including health and medical area. At the same time, Wireless Sensor Networks(WSNs) has some new development situations. In real world, people’s health status detection/prediction is influenced by various factors such as social relationships, history health statuses and people’s personal condition. However, few publications systematically study how health statuses evolve in a dynamic social network and to what extent different factors affect the user health status. In this paper, we first describe a novel Medical Social Networks(MSNs) which is a classic kind of Medical Internet of Things(Medical IoTs). Then combining these above factors together, we propose a unified model, namely TS-FGM, based on Probability Factor Graph Model, and thus present a novel health status prediction method based on TS-FGM in MSNs. More specifically, users’ health statuses at time t are influenced by their private attributes, their own health statuses at time t-1 and their neighbors’ health statuses at both time t and t-1. At last, we present an efficient decision-fusion-oritented algorithm to learn the model. Finally, we validate the model on real-world data sets in Twitter. And we compare our method with baseline algorithm SVM on a real clinic medical data set for pulse diagnosis. Experimental results show that the model is effective and the proposed method partly outperforms the baseline method for disease prediction.