A Framework for Mapping User Designed Forms to Relational Databases
In the quest for database usability, several applications enable users to design custom forms using a graphical interface, and forward engineer the forms into new databases. The path-breaking aspect of such applications is that users are completely shielded from the technicalities of database creation. Despite this innovation, the process of automatically integrating a new form into an existing database remains unexplored. At large, databases continue to remain unusable. This dissertation focuses on investigating the problem of mapping multiple forms into an existing relational database. We seek a framework that automatically detects and merges the semantically matching elements between forms and databases. Upon encountering the unmatched form elements, the framework creates new database elements and integrates them with the underlying database. The technical goal is to ensure that the resultant database is compliant with "high quality" principles defined in terms of form semantics. The usability goal is to ensure minimalism in the user interventions required to discover correspondences.
We introduce a model, the form tree, to represent the user's semantic intentions represented by a form. We design two anchor approaches to extract the form tree from an arbitrarily-designed form, and to further disambiguate the form semantics by annotating its terms using standard concepts. Thereon, we formulate the following mapping solution: (i) Leverage linguistics and semantics to discover and validate semantic correspondences between the form tree and the existing database. (ii) Devise mapping algorithms that create a new high-quality database for a given form tree, and merge it with the existing database while fusing the semantically equivalent elements. We evaluate the entire framework by developing a prototype, and experimenting in the healthcare domain. We collect 52 clinical forms from different medical institutions, and map them to 6 databases of varying scales. The anchor approaches generate form trees with 98% accuracy, and annotate the form terms with a precision of 0.89. The framework helps in producing up to 74% principle compliant databases in terms of compactness, and in reducing user interventions by 61%. The experiment results imply that the use of annotation helps in improving the quality of the evolved database.
Dr Yuan An
Winner, Most Outstanding Promise Doctoral Award, Drexel University Commencement Day, June 2012
Nominated, Best Doctoral Dissertation Award, iConference 2013