Patient similarity using network structure properties in online communities

Abstract

The growing amount of user generated content in healthcare social media requires new methods to gain new insights about patients as users of online health communities. One interesting general problem is filtering information and making relevant information more conspicuous to users in Internet websites. Previously, collaborative filtering techniques have been successfully applied in recommender systems for personalizing Internet shopping and movie portals. Current work uses network structures - inferred from patient interactions - to enhance patient-similarity analysis, for predicting the top-N threads in online communities. Using network structure properties unique to healthcare social networks, experimental results based on the Euclidean distance, Pearson correlation and Tanimoto similarity, confirm that community structure properties can enhance recommendations. The results are comparable to other collaborative filtering methods proposed in the literature. These findings have implications for designing personalized health-related social media.