Inductive Databases

Abstract

Data mining aims at trying to locate interesting patterns or regularities from large masses of data. Data mining can be viewed as part of a data analysis or knowledge management. In data analysis tasks one can see a continuous spectrum of information needs, starting from very simple database queries (“what is the address of customer NN”), moving to more complex aggregate information (“what are the sales by product groups and regions”) to data mining type of queries (“give me interesting trends on sales”). This suggests that it is useful to view data mining as querying the theory of the database, i.e., the set of sentences that are true in the database. An inductive database is a database that conceptually contains in addition to normal data also all the generalizations of the data from a given language of descriptors. Inductive databases can be viewed as analogues to deductive databases: deductive databases conceptually contain all the facts derivable from the data and the rules. In this talk I describe a formal framework for inductive databases and discuss some theoretical and practical problems in the area.