Inductive Databases as Ranking
In: DaWaK 2004, 1-3 Sep 2004, Zaragoza, Spain.
Most of the research in data mining has been focused on developing novel algorithms for specific data mining tasks. However, finding the theoretical foundations of data mining has recently been recognized to be even greater concern to data mining.
One promising candidate to form a solid basis for data mining is known as inductive databases. The inductive databases are databases with a tight integration to data mining facilities. However, it is not clear what inductive databases actually are, what they should be and whether inductive databases differ notably from the usual databases with slightly broader notions of queries and data objects.
In this paper we aim to show that the viewpoint offered by inductive databases differs from the usual databases: the inductive databases can be seen as databases with ability to rank data manipulation operations. We describe how several central data mining tasks can be naturally defined by this approach and show that the proposed inductive databases framework offers conceptual benefits by clarifying and unifying the central data mining tasks. We also discuss some challenges of inductive databases based on query ranking and grading.