Integration of multiple Internet sources poses many novel challenges such as handling semistructured data, creating views without complete knowledge of the source structure, adapting to the limited query interfaces (capabilities) of the sources, and answering queries referring to virtual objects which are the result of complex information fusion operations. Recent works have provided flexible, though very often inefficient, ways to rapidly develop mediators that achieve the above goals. This research develops data structures, algorithms and systems which can enhance the performance, often by orders of magnitude, and improve the user interface without compromising the development flexibility. Data guides are proposed for describing the concepts (or combinations of concepts) that significantly appear in each source and in the integrated view created by the mediator. Data guide computation algorithms use (1) the mediator specification, (2) the parser of the source data, and (3) actual data of query results. Mediators and user interfaces are extended to make use of the data guides upon query processing and query formulation. This research also develops algorithms that (1) further extend the mediator's ability to adapt to the limited capabilities of the sources, (2) use knowledge (e.g., dependencies) of the source data to enhance the performance of object fusion queries, and (3) optimize the delivery of partial answers. This research will demonstrate how flexible integration systems (such as systems for constructing virtual malls by integrating multiple Web shopping sites or systems for fusing all available financial information relating to a company) can be rapidly developed while still achieving the performance of hard-to-built custom made integration systems.
