When querying a database, a user is restricted to a given and frozen organization of information, enforced by the creator when designing the schema. Should the user send a request beyond the schema, he will be denied access to the expected information. The creator has imposed his view to the user. This is the limitation of the source-driven approach.
Our user-oriented paradigm grants more flexibility by allowing the user to design his own view on demand. There is no magic: the limits are transfered to the extraction capabilities of the system. AKIRA's administrator is in charge of representing these capabilities at a conceptual level in terms of schema components, in a modular way. He provides the user with a pool of schema components that can be combined to specify the user's view.
We consider three extraction capabilities: extracting concepts, meta-concepts and attributes which we represent in an abstract way in a pool of schema components. Extracting a concept goes beyond the usual retrieval of keywords, even though the extracting technique may use a list of keywords. We suppose that a concept is extracted when the following steps are accomplished: (1) recognition, and (2) identification (canonical representation). The recognition phase consists in selecting fragments of documents which satisfy a certain criteria. The retrieved fragments are identified in a second step. An IE tool capable of recognizing and identifying conference names is represented by a concept class Conference with attribute name. Similarly, a date extraction tool corresponds to a class Date with attributes month, day and year. Each of these classes is a schema component by itself. A concept class can be specialized according to other extraction capabilities. For example, attribute topic, listing topics of interest, can specialize class Conference. Two concept classes can be also combined through a meta-concept such as submission_deadline to assemble a new conceptual schema.
Copyright AKIRA Project, University
of Pennsylvania
Department of Computer
and Information Science, Institute
for Research in Cognitive Science