DEDICATION To my great - grandfather Shri

OF THE DISSERTATION Personalized Situation Recognition By Vivek Kumar Singh Doctor of Philosophy in Information and Computer Science University of California, Irvine, 2012 Professor Ramesh Jain, Chair With the growth in internet-of-things, social media, mobile devices, and planetaryscale sensing, there is an unprecedented opportunity to assimilate spatio-temporally distributed streams into actionable situations. Detecting situations in realtime can be used to benefit human lives and resources in multiple applications. However, the progress in the field of situation recognition, is still sluggish because: (a) the notion of situations is still vague and ill-defined, (b) there is a lack of abstractions and techniques to help users model their situations of interest, and (c) there is a lack of computational tools to rapidly implement, refine, and personalize these situation models to build various situation-based applications. This dissertation computationally defines situations and presents a framework for personalized situation recognition by providing support for conceptual situation modeling, data unification, real-time situation recognition, personalization, and actiontaking. The proposed framework defines a situation as “An actionable abstraction of observed spatio-temporal descriptors”, and identifies a data representation, a set of analysis operations, and lays out a workflow for modeling different situations of interest. Considering Space and Time as the unifying axes, it represents data in a grid-based Exv mage data structure. It defines an algebra of operations (viz. Selection, Aggregation, Classification, Spatio-temporal Characterization, and Spatio-temporal Pattern Matching) for situation recognition; and defines a step-by-step guide to help domain experts model their situations based on the data, the operations, and the transformations. The framework is operationalized via EventShop a web based system which lets users graphically select, import, combine, and operate, on real-time data streams to recognize situations for generating appropriate information and actions. EventShop allows different designers to quickly configure their situation models, evaluate the results, and refine the models until a satisfactory level of performance for supporting various applications is achieved. The detected situations can also be personalized and used for undertaking control actions via Situation-Action rule templates. The framework has been used to build multiple applications including flu monitoring and alerting, wildfire recognition, business decision making, flood alerts, asthma recommendation system, seasonal characteristics analysis, and hurricane monitoring. Thesis statement: We computationally define the notion of situations and present a framework for personalized situation recognition by providing support for conceptual situation modeling, data unification, real-time situation evaluation, personalization, and actiontaking.