Novel Computational Approaches to Information Retrieval and Data Mining
The realities and opportunities of the global information environment enable and necessitate new techniques and approaches, they expand the scope and the methodology of database theory. This talk surveys recent results of this nature, by the author and/or several collaborators, in two areas. In information retrieval, spectral methods have been introduced which extract the hidden semantics of a corpus by analysing the eigenvalues of related matrices. In fact, the performance of such methods can be theoretically predicted to be favorable in a certain statistical sense. A different spectral method has also been introduced successfully in the analysis of hypertext so as to identify authoritative sources of information. In data mining —the search for interesting patterns in data— we argue that a meaningful definition of “interesting” requires consideration of the optimization problem the enterprise is facing. This “microeconomic” view leads quickly to certain novel and interesting computational problems.