Discovering Non-Standard Semantics of Semi-Stable Attributes

A new class of rules, called action rules, show what actions should be taken to improve the profitability of customers. Action rules introduced by (Ras and Wieczorkowska, 2000a) and investigated further by (Ras and Gupta, 2002a) assume that attributes in a database are divided into two groups: stable and flexible. These reflect the ability of a business user to influence and control their change for a given consumer. In this paper, we introduce a new classification of attributes partitioning them into stable, semi-stable, and flexible. Values of stable attributes can not be changed for a given consumer (for instance "maiden name" is an example of such an attribute). So, stable attributes have only one interpretation. If values of an attribute change in a deterministic way as a function of time (for instance values of attribute "age" or "height"), we call them semi-stable. All remaining attributes are called flexible. Clearly, in the process of action rule extraction, stable attributes are highly undesirable. What about semi-stable attributes? Although, they seem to be quite similar to stable attributes, the difference between them is quite essential. Semi-stable attribute may have many different interpretations but among them only one interpretation is natural and it is called standard. All its other interpretations are called non-standard. In a non-standard interpretation, a semi-stable attribute can be classified as flexible. In a single database we may easily fail to identify attributes which have non-standard interpretation. In this paper, we show how distributed information system introduced by (Ras, 2000b, 2001a) can be used as a tool to identify which semi-stable attributes have nonstandard interpretation so they can be classified as flexible. This way, by decreasing the number of stable attributes in a database we may discover action rules which would not be discovered otherwise.