Projects: Contextual Computing

The goal of user modeling is to produce highly relevant information ecologies on a per user basis. While simple in principal, past research suffers from a number of issues. First, instrumentation of work environments, applications, and other processes is difficult. Without a theoretical approach, decisions about what user data to collect and how to collect it are problematic and have limited past modeling efforts. Many systems fail to capture highly informative data streams while remaining unobtrusive and addressing privacy concerns. Second, many models are not robust, either as a result of modeling user interactions within a limited number of application, or focusing on shallow data like preferences, demographics, etc. This typically results in brittle, one-dimensional models. Finally, most systems require explicit user input to produce the initial user profile, creating a system that significant up-front manual labor to be effective.

Leveraging our existing technologies in instrumentation, content analysis, linguistics, and task analysis, we propose to develop a user-modeling infrastructure that tackles the challenging problems of modeling content consumption as well as sensemaking tasks. The user modeling research has four primary components that address the above three problems:

a) Analyst modeling. (To tackle the first instrumentation problem.) We decide what user data to collect based on cognitive theories of work practices. Using the theories of Information Foraging and Sensemaking, the modeling of analysts will focus on their content consumption and their sensemaking strategies. Interfaces will capture data streams and artifacts of external cognition collected from the instrumentation layer as well as from prototype applications.

b) Semantic analysis of sources. (To tackle the second robustness problem.) Strong modeling methods will require more semantic understanding of the source material. Our prior research into the personalization of information access demonstrated the utility of mapping sources onto a canonical ontology. We propose to extend that research by incorporating other dimensions, using higher-level semantic representations of sources like topics, concepts, point-of-view, etc. as well as other dimensions/attributes of the source material.

c) Management of user profiles. (To tackle the third user effort problem.) Creation of user profiles will be affected by our understanding of what works best for analyst modeling, while paying particular attention to what methods require least user effort. Analysts will be able to inspect, modify, compare, and share user profiles and adapt them as necessary.

d) Adaptation of the information ecology. (To tackle the integration issues.) To test out the above solutions, the proposed prototypes will be able to contextualize and personalize content and services. So the user modeling system provides a vital repository able to communicate artifacts gathered from different stages of sensemaking to inform decisions in other stages.