Virtual Answers for Query Reenement in Information Retrieval

Query reenement in information retrieval over repositories on the Web faces problems when trying to adapt a purist database approach. Needed semantics and brokering facilities diier signiicantly when compared, for instance, with traditional SQL-like query schemes. Therefore, we propose an interaction model, adapted to Web sources, and based on constraints and on the notion of so-called \virtual answers". Our model extends both the basic pull and push models, but proposes virtual answers as a powerful means for further interactions from a given answer: typically, query reenement. A major contribution of this paper is the discussion of the issues of information combination (brokering) and of soundness and completeness of information retrieval engines (semantics) in the framework of this new model. We make the simplifying assumption that sources evolve monotonically. Strictly speaking, this may look irrealistic in the context of the Web, but it approximates many real situations, where changes in the sources occur on a larger timescale than user sessions, or where clients check the validity of returned items before actually making use of them. Currently, the model is being evaluated for integration into the Constraint Based Knowledge Broker system (CBKB 1). The CBKB system aims at heterogeneous information retrieval, schema integration, and knowledge fusion.