Extraction Of Health Information From Multiple Data Sources

With the emergence of Internet,the space for human interaction is expanding from the traditional physical world into cyber space,which helps human to address the restrictions of physical world,boosts the information dissemination and enhances the user interaction.The advantage of unprecendented sensing capability and the embracing of smart mobile terminal enable us to seamleassly collect the digital records and boost the emergence of Cyber-Physical-Social System(CPSS).However,the challenges of building one CPSS application mainly lie in the heterogeneity of sensors,the diversity of data sources,and the dynamic of user interaction.The information extraction from multiple data sources is important for the CPSS applications.To overcome these challenges,this dissertation focuses on the extraction of health information from multiple data sources including sensing data and online user-generated data.The main work and contributions in this dissertation are shown below:(1)Modeling of Multiple Data SourcesThe data in CPSS is hetereneous,dynamic,sparse and noisy.To address these challenges,we classify the data in the CPSS into two types(sensing data and user-generated data)according to the generation mechanism,and define two data models respectively.Based on the data models,we elaborate the framework of data fusion to extract high-level information from the multiple data sources.Due to the heterogeneity and diversity of data,ontology modeling is introduced to describe the relationships among different metadata,and the rule-based reasoning is utilized to extract high-level knowledge from heterogeneous data.(2)Extraction of Health Information from Sensing DataMany sensors are utilized to detect vital signs,which makes the long-term and seamless monitoring of health condition feasible.This dissertation focuses on the evaluation of calorie consumption based on user daily activity using the single acceleroate in the smart phone.First,a comprehensive overview about activity recognition is conducted to find the loopholes in the existing research work.Due to the features of resource-restricted devices,a hirerarchical light-weight algorithm is proposed,which works well for the low resolution data and reduces the energy consumption of mobile devices.Finally,the metabolic equivalents are utilized to quantify the calorie consumption based on user daily activities.(3)Extraction of Health Information from User Generated DataUsers are using social media as one important platform to retrieve and share health-related data.However,the ambiguity and sparsity of online data make it difficult to find health-related content for individual users.This dissertation focuses on how to mine substance useage based on the online data from individual users.First,a variety of patterns were extracted from different prespectives including social network analysis,text mining,sentiment analysis,and information theory.Then,we propose a deep learning based algorithm to predict whether the users are smokers based on the user-generated content.(4)Data Fusion from Multiple Data SourcesDue to the features of data in CPSS,this disertaion focuses on the issue of data fusion in CPSS,and investigates the data fusion from two different levels: feature level fusion and decision level fusion.For the decision level fusion,a comprehensive ontology model is designed to define different contexts including person,device,environment,network and information.The rule-based reasoning can integrate multiple data sources and infer high-level knowledge.For the feature level fusion,a user similarity model is proposed based on social interaction and topics.Based on the similarity model,a clustering algorithm is proposed to find appropriate community members with strong interaction or similary topics.(5)Application and EvalutionMany elder adults have poor medication adherence,which may result in adverse drug events,significant morbidity and mortality.To address this problem,the Socialized Prompting System(SPS)is proposed,which combines ubiquitous sensors in the smart space and mobile social networks to improve medication adherence.Ubiquitous sensors facilitate the seamless monitoring of medication intaking behaviors.Meanwhile the mobile social networks contribute to social prompting in a community.One prototype is implemented using the OSGi framework,and preliminatary evaluation is conducted to demonstrate the feasibility of the proposed system.