Data Cognition

In Dallas, the author of a database textbook is working on the next edition of his book. He is considering the chapter on SQL. What small extract of this (massive) language should be emphasized? Which constructs might students have particular difficulty understanding or using in practice?

In Washington, DC recently, in response to thousands of product recalls, including the lead toy fiasco (see reference), and the associated difficulty of tracking every product deemed a threat to the health and safety of consumers, the federal government responded with the Consumer Product Safety Improvement Act (CPSIA) of 2008. In accordance with the Act, the Consumer Product Safety Commission (CPSC) is developing a publicly accessible, searchable online database of consumer product incident reports (see reference). The experts at CPSC realize that if one product is found to be defective another may need to be recalled as well and are grappling with issues related to data semantics. How can semantics among objects be defined transitively and, perhaps, recursively? How to establish temporal and spatial relationships, i.e., temporal and spatial semantics, between risk factor exposure and health events around the country?

In Boston, the next major release of a database configuration tool is being specified. The challenge is to mediate between the seemingly contradictory goals of increased functionality and unnecessary instruction. The designers ask: Can the customers contend with the additional complexity that generally accompanies greater flexibility? Do the configurations that are created with this tool correspond with the customer's mental model?

Perhaps also at this minute in Salt Lake City a developer is working on an iPhone application that will enable a user, via touch-screen clicks and a highly optimized (and quite circumscribed) display, to navigate extensive data on climate, soil chemistry, plant varietals, and personal goals to arrive at a recommendation of precisely the right annuals and fertilizer for the envisioned amateur gardener to purchase at Home Depot. Does this gardener need a mental list of the factors that impact such a decision? Should the system direct the querying or should it be open to the user? How does the user understand the results of the query?

While the scenarios described above are disparate, the underlying question common to SQL standards committee members, database textbook authors, implementers of online databases, database tool vendors, and data-intensive application developers is:

Conceptual modeling and query languages are critical to organization and utilization of data. While there has been a considerable amount of past work in designing these models and languages (primarily by commercial vendors of such tools), and some research in understanding them, there exists no coherent, comprehensive user model of data semantics and access. Such a model has not been developed in part because different communities have utilized different research methodologies in a fragmented way.

We are articulating and validating a theory-based user model of data semantics and of data access that can be used to better organize and utilize data. This model can also be used to evaluate existing conceptual data modeling and query languages and to design future modeling and query languages more effectively. To achieve this objective, this project integrates perspectives and methodologies from both the computer science (CS) and management information systems (MIS) research disciplines.

Integrating and testing in a scientific manner a comprehensive user model of data semantics and access requires a broad program of research spanning diverse activities ranging from eliciting domain knowledge from users, to modeling the domain, to designing databases using principles that reflect domain semantics, to examining how to effectively map those designs to relational schemas, to studying the use of query languages to extract required information from these databases. These activities are linked by common requirements to understand domain semantics in a fundamental way, to preserve semantics in a series of transformations from a mental model to a machine-executable representation, and to retrieve information in a way consistent with users' understanding of the domain and the models used to represent it.