Data Mining Opportunities in Very Large Object Oriented Databases

Information overload can hamper extracting hidden knowledge from a database. Data mining techniques ooer automated exploratory data analysis of databases. The mining process reveals knowledge buried in the data and provides insights into these data. Conventional mining techniques such as neural nets, regression analysis, and the discovering of rules uncover hidden information from user-provided training examples. A user indicates several of objects which are of interest to him (positive training examples) and few objects which are of no interest to him (negative training examples). The system then automatically discovers all interesting objects. In this position paper we present data mining opportunities that arise in very large object oriented databases, wherein the users no longer need to provide training examples. Data mining is achieved by user set priorities, user speciied guessed queries, and the notion of complete known information (i.e. information closure) about an object. From these inputs, hidden knowledge buried among huge amounts of information can be extracted in a user-driven manner.