User Defined Aggregates for Logical Data Languages

A new wave of data-intensive and knowledge-based applications|such as data mining and decision support|require the introduction of complex application-speciic aggregate functions. In this paper, we propose extensions for deductive database systems to support these new applications. We develop constructs, formal semantics, and implementation techniques for user-deened aggregates, and describe their realization in an extended LDL++ system recently built at UCLA. With these extensions, the system can support on-line aggregation, roll-ups for data cubing, temporal aggregates for time-series, iceberg queries, and other recently proposed operators used in decision support and data mining procedures. We then discuss the application of this technology to other DBMSs, and in particular to the SQL3 speciications that support the notion of user-deened aggregates. We show that SQL3 suuers from limitations that severely restrict its use in new applications; thus we propose simple extensions similar to those used for LDL++ to overcome such limitations.