Mining Databases and Data Streams with Query Languages and Rules

Among data-intensive applications that are beyond the reach of traditional Data Base Management Systems (DBMS), data mining stands out because of practical importance and the complexity of the research problems that must be solved before the vision of Inductive DBMS can become a reality. In this paper, we first discuss technical developments that have occurred since the very notion of Inductive DBMS emerged as a result of the seminal papers authored by Imielinski and Mannila a decade ago. The research progress achieved since then can be subdivided into three main problem subareas as follows: (i) language (ii) optimization, and (iii) representation. We discuss the problems in these three areas and the different approaches to Inductive DBMS that are made possible by recent technical advances. Then, we pursue a languagecentric solution, and introduce simple SQL extensions that have proven very effective at supporting data mining. Finally, we turn our attention to the related problem of supporting data stream mining using Data Stream Management Systems (DSMS) and introduce the notion of Inductive DSMS. In addition to continuous query languages, DSMS provide support for synopses, sampling, load shedding, and other built-in functions that are needed for data stream mining. Moreover, we show that Inductive DSMS can be achieved by generalizing DSMS to assure that their continuous query languages support efficiently data stream mining applications. Thus, DSMS extended with inductive capabilities will provide a uniquely supportive environment for data stream mining applications.