Understanding and Utilizing the Influence of SocialNetworks on Health Care in YesiWellDejing Dou, Daniel Lowd, Jessica Greene, Brigitte Piniewski, Ruoming JinXintao Wu, David Kil, and Frances ShinDecember 15, 2010Motivation: Two thirds of the US population are now overweight or obese. This incurs signifi-cant health risks and financial costs to society. Obesity and overweight, although multifactorial, canlargely be explained by the social and cultural spread of poor health habits such as lack of exercise,smoking, fast food consumption, and alcoholism spread through communities, introducing and re-inforcing harmful behaviors [1]. By the time behaviors develop into diseases and individuals seekmedical help, it is often too late to reverse the chronic poor health outcomes. Traditionally, supportgroups and other social reinforcement approaches have been popular and effective in dealing withunhealthy behaviors including overweight. Of the factors associated with sustained weight loss oneof the most important is continued intervention with frequent social contacts.Recent advances in mobile technology and online social networks provide new opportunitiesto support healthy behaviors through lifestyle monitoring and online communities. The YesiWellproject1uses both approaches together to help people maintain active lifestyles and lose weight.Though still in the pilot stages, YesiWell already has a wealth of data on the health statistics, ac-tivity levels, and social network activity of a large group of users. YesiWell is expected to have up to2 million users in next couple of years. The YesiWell data specifically include three dimensions: i)physical activities such as the aerobic steps, regular steps, speed, and duration. Each user carriesa sensor called hPod which can do physical activity tracking as well as aggregation of weight andblood pressure data. ii) biometrics and biomarkers such as serum creatinine, total cholesterol, low-density lipoprotein (LDL), high-density lipoprotein (HDL) etc. iii) social activities tracked throughuser-generated social networks, and provides social support, discussions, and competitions.Background: To exploit the full potential of wellness systems and social networks such as Yesi-Well, it will be necessary to develop new methods in data mining, social network analysis, knowledgerepresentation, and privacy that can handle the complexities of data that is high-dimensional, largescale, temporal, social, and very sensitive. The authors of this white paper propose a novel frameworkto address this need. The research team include the experts in data mining, social networks, healthpolicy, ontology, privacy, and health informatics. The first two authors are potential participants of theworkshop if the paper is invited. Dejing Dou is an Associate Professor in Computer and InformationScience (CIS) at the University of Oregon. Professor Dou is an expert in ontology-based data miningand data integration. He is the director of the Advanced Integration and Mining Laboratory. ProfessorDou is specially interested in applying Semantic Web ontologies and data mining into biomedical1http://www.yesiwell.com/1informatics and health informatics. He is currently the PI of a NIH funded four year R01 project titledwith Neural ElectroMagnetic Ontologies (NEMO): ERP Knowledge Representation and Integration.Since the spring of 2010, Professor Dou has been collaborating with Dr. Brigitte Piniewski (PI ofYesiWell), Prof. Jessica Greene (expert in health policy), Mr. David Kil (Chief Scientist of Yesi-Well) for data warehousing in YesiWell. Daniel Lowd is an Assistant Professor in the Department ofComputer and Information Science at the University of Oregon. His research covers a range of topicsin statistical machine learning, including statistical relational representations, unifying learning andinference, and adversarial machine learning applications (e.g., spam filtering). Prof. Lowd have beencollaborating with Prof. Dou, Prof. Ruoming Jin (expert in social network mining), and Prof. XintaoWu (expert in privacy preserving mining) for YesiWell data mining.Vision: Our vision is to develop new methods in a variety of areas, building on the current state-of-the-art. We will utilize temporal Granger causality and dynamic Bayesian networks to study theinfluence of social networks on health care. We will design a formal semantic model with SemanticWeb ontologies including common concepts and mappings for health behaviors, biomarker measures,and social activities, and a sophisticated way to reuse those ontologies to facilitate data mining andsocial network analysis. We will also develop novel differential privacy preserving techniques to pro-vide strict privacy guarantees during the process of collecting, querying, analyzing, and mining suchcomplex health data with social networks. Finally, we will evaluate our data mining algorithms bothby the quantitative measures and the tailored communication from the social science point of view.Though we will focus on mining and modeling the real-world data from the YesiWell environment, theapproaches, methodology, and results developed and discovered in the projects can be applied in anyother emerging healthcare social networks. Given the widespread use of personnel health systems andthe research on the influence of social networks, development of such data mining algorithms, socialnetwork design, ontologies, and privacy preserving methods is both timely and significant.Internet-based health interventions hold particular promise for studying the impact of social net-works on health care, especially for tailored communication. While Internet-based tailored healthcommunications are widely used, the research in this area is still quite limited. Our research willhelp overcome these limitation by leveraging sophisticated data mining and ontology techniques. Afurther benefit will be the support for mining other cross-dimensional and complex network data. Theprospect of data mining from different dimensional and large scale network data is exciting, thoughchallenging for a variety of reasons. Many researchers would agree that multi-dimensional miningand social network analysis is one of the major aims for advancement of health informatics.Also, understanding the influence of social networks on health activities could have clinical, aswell as basic research, applications. For example, using the YesiWell data, will these new data miningtools help separate the