PhD Defense: Introducing Dynamic Ontologies to Encode and Manage Relevance in Context Aware Systems
Context aware systems, to date, tend to fall into one of two categories: domain specific or generic across multiple domains. Domain specific systems are single-use instances – that is, establishing the ability to manage context for an additional domain necessitates the creation of an additional system. Authors of such systems should instead strive for generic ones. Generic context management systems require a generic modeling and context delivery system.
Previous research has shown that generic context aware systems prove to be quite dynamic through their use of ontologies. These ontologies, however, are very rigid in nature, requiring additional software to mature and manage instantiated models, filter relevant information, or pre-cache information. The result is users who wish to use generic systems must encode relevance across ontological models, filters, and newly created external software with each re-use in order to manage context manipulation at run time.Through the design and implementation of Rover3, while leveraging the concept of an Automatic and Dynamic Information Model (ADIM) methodology, we outline what we believe how context aware systems should function. By providing a framework to encode relevance within ontologies, we minimize the way to present and consume relevant information. Our context management framework uses dynamic ontologies to deliver relevant information to users striving to achieve goals for any given situation.Walking through an accident response case study we showcase the aforementioned features of Rover3, showing how such incidents can benefit from context aware systems. The value of Rover3 is expressed through an extensibility study where efforts to expand existing ontological models are compared between Rover2 and Rover3.This dissertation presents:
The notion of relevant context and how it can be managed at runtime through a generic context aware system. We do so by explaining what it means for relevant context to expand and contract as information is gathered.
The required primitives and rules for modeling any generic situation. We accomplish this through an updated context model ontology, poised to better represent generic information and better serve generic context aware systems.
The Automatic and Dynamic Information Model (ADIM) methodology, how one can encode relevance in a general information model, and exhaustive grammar and rules for this version of ADIM. With ADIM, system designers are able to encode relevant context into a generic model and define the rules for context expansion/contraction at run-time.
The Rover3 system and its application of ADIM, showcasing how it provides a generic framework to model and manage context that does not require any additional software. We exhibit this through a lengthy case study and an evaluation against an existing context management system.
Chair: Dr. Ashok Agrawala Dean's rep: Dr. Min Wu Members: Dr. Alan Sussman Dr. Atif Memon Dr. Adam Porter