Invited "Workshop-closing Talk" by Arno Siebes

Where is the mining in KDID?

From a database point of view, KDID means that data mining becomes an integral part of the database. For example, mining algorithms become a part of SQL, the results can be stored in the database and subsequently queried. There has been lots of work on this, this workshop is an excellent example of this.

However, from a data mining point of view, this is only the beginning and not necessarily the most interesting part. A data miner wants to mine. If the database contains models and patterns, the data miner wants to mine models and patterns. There has been far less focussed research on this aspect. This is a shame, because if we do not treat models and patterns as just another kind of data, how can we convince others that that models and patterns have to be first class citizens in databases?

Part of the reason for this lack of concerted effort is probably that it seems not that easy to think of good applications. In this talk I will argue (and illustrate) that this impression is wrong. Pattern mining is a kind of feature extraction, something we routinely do in pre-processing. In other words, mining patterns is no different from standard data mining.