This chapter gives a short description of a prototype Decision Support System (DSS), which assesses the value and/or utility functions of the individual user. This DSS allows de-facto training of the computer in the same preferences as that of the individual user without the need of additional participant or mediator in the process of utility evaluation. It is mathematically backed up by the methods described in the preceding chapters. The presented methodology and mathematical procedures allow for the creation of such individually oriented DSS for analytic representation of the preferences as objective function based on direct comparisons or on the gambling approach. Such systems may be autonomous or parts of a larger information decision support system.
Top1. Data Mining And Knowledge Discovery
Data mining generates information about situation structure. Usually, the situation is interpreted to mean a problem, system, or condition. The systematic observation of a situation requires its consideration in certain framework that gives a generic view of its elements and system-based description. The framework guarantees representation and record of situation appearances. A system consists of interrelated objects. An object can be either a real entity or a mere concept (Andreev, 2001). It comprises a set of attributes leading to a precise definition of the system as a relation among attributes of objects. The value of an attribute may be either variable or constant.
An appearance of a situation is described by the knowledge about the system represented through objects’ attributes. A situation’s appearance is given by the values that its constituting attributes hold, when the situation is observed at a particular point in time t. Change of values of single attributes results in a modification of the situation’s appearance. A sequence of appearances of a situation presents situation (system) behavior. The latter is restricted by the domains of values of the system attributes. However, the situation behavior does not generally imply subsequent appearances to depend on each other. The situation structure is determined by means of static relations and dynamic relations. The former represent constraints on the values of single attributes and interdependences among the attributes of an appearance. Dynamic relations specify the interdependence of the values of attributes regarding subsequent situation’s appearances.
Observation of situation behavior is an elementary way of gaining knowledge about a situation. Recording of situation’s appearances provides data representing this knowledge. However, situation determination (formulation) requires knowing of values of situation’s attributes beyond the data available from observation. It might either not be possible to observe the values of certain attributes of a situation’s appearance at hand or the value of a single attribute of a future appearance might be unknown. In both cases information about situation structure is needed for inference unknown attributes’ values. Information about static relations allows for drawing conclusions on the value of an attribute on the base of the values of other attributes of the same situation’s appearance. The dynamic relations give information necessary for making forecasts about values of attributes of future appearances.
Data mining is an information extraction activity that could be presented by actions with different goals. Such a goal is to search information that is necessary for problem formulation. This information could be derived not only from some information system, but from experts familiar with the situation, as well. Another goal that determines another type of action realizing data mining is to search large volumes of data for patterns and discover hidden facts contained in databases. A unique definition of data mining has not been established yet, since its meaning is not determined by the objective of an action. This activity can be realized by various actions.
From a scientific viewpoint, the data mining process is regarding as data evaluation that consists of two steps (actions): a preprocessing step and core data mining action that seek for a representation of either static or dynamic relations of a situation. It is concerned with secondary analysis of a huge amount of data typically being at hand in the form of recorded situation’s appearances. The data is not collected based on experiments designed to answer a certain set of a priory known questions. It is result of either a situation simulation or actual acting of a situation. However, the data cannot be processed by data mining activity without hypothesis about situation (system) structure. Such an a priori hypothesis determines the type of information to be derived from data. The preprocessing step is denoted as a map from a set of recorded appearances into a target data set (Meisel, 2007). This map is implemented by various instances of the following types of preprocessing operations—selection of a subset of appearances, projection of situation’s attributes into a subset of these attributes, modification of the domain of an attribute, construction of new attributes by aggregation of original attributes. Usually the preprocessing operations are implemented as search procedure that ensures an optimal target data set with respect to the performance criteria of the subsequent core data mining action.