Inductive Querying with Virtual Mining Views

Hendrik Blockeel (1) , Toon Calders , Elisa Fromont (2) , Bart Goethals (3) , Adriana Prado (2) , Céline Robardet (4)

1 DTAI - Declarative Languages and Artificial Intelligence
2 LHC - Laboratoire Hubert Curien
3 Adrem - Advanced Database Research and Modelling
4 DM2L - Data Mining and Machine Learning

An important motivation for the development of inductive databases and query languages for data mining is that such an approach will increase the flexibility with which data mining can be performed. By integrating data mining more closely into a database querying framework, separate steps such as data preprocessing, data mining, and postprocessing of the results, can all be handled using one query language. In this chapter, we compare 6 existing data mining query languages, all extensions of the standard relational query language SQL, from this point of view: how flexible are they with respect to the tasks they can be used for, and how easily can those tasks be performed? We verify whether and how these languages can be used to perform four prototypical data mining tasks in the domain of itemset and associa- tion rule mining, and summarize their stronger and weaker points. Besides offering a comparative evaluation of different data mining query languages, this chapter also provides a motivation for the next chapter, where a deeper integration of data mining into databases is proposed, one that does not rely on the development of a new query language, but where the structure of the database itself is extended.

Mots clés: Inductive databases; Constraint extraction; Data mining

Domaines: Intelligence artificielle [cs.AI]

Hendrik Blockeel, Toon Calders, Elisa Fromont, Bart Goethals, Adriana Prado, et al.. Inductive Querying with Virtual Mining Views. Džeroski, Sašo; Goethals, Bart; Panov, Panče (Eds.). Inductive Databases and Constraint-Based Data Mining, Springer, pp.265-288, 2010. ⟨hal-00531170⟩