Decisions in the modern world are often made in rapidly changing, sometimes unexpected, situations. Such situations require availability of systems / tools allowing fast and clear description of situation, generation of new and reuse of previously made effective solutions for situation reformation, selection of a right decision maker and supplying him/her with necessary data. Such tools include components for actual situation description, user modeling, finding appropriate methods for problem solving, integration of data from heterogeneous sources, finding / generation of insufficient data, removing uncertainties, estimating solutions, etc. During decision making process a large amount of auxiliary raw data are accumulated in repositories. Methods of data mining are used in such systems for different purposes: finding associative rules between decisions and factors affecting them, user clustering using decision trees and neural networks, recognition of common users’ features / interests and others (Chiang et al., 2006; Li, 2005; Thomassey and Fiordaliso, 2006). Validation of the obtained results can be performed using simulation software modules.
TopBackground
The chapter presents a developed approach that assumes usage of (i) ontologies for application domain description (notions, relations between notions, data sources and methods), (ii) user profiles to accumulate raw data and build the system’s vision of the user, and (iii) context for actual situation description. The developed approach is oriented to producing an ontology-driven context model so that the decision makers would be provided with the information and knowledge required in the situation they are interested in and according to the roles they play.
Within the presented approach the context is a weakly-structured information containing three constituents: (i) ontology elements describing the actual situation; (ii) user data representing user’s role, preferences, competences, etc.; and (iii) data / information / knowledge extracted from available sources and relevant the actual situation. Context is built when a situation occurs and used for processing requests related to this situation. The context is a basis for generation and estimation of alternative solutions and presenting them to the decision maker for selecting the best one from his/her point of view. Each situation is described by the following components: context, solutions generated using the context and the final decision (selected solution / solutions). These components are stored in the user profile.
Finding influence of the context on the final decision made by the decision maker playing a certain role is an important task because it helps to (i) find and model typical scenarios of interaction between users; (ii) reveal typical situations within large amount of raw data; and (iii) cluster existing decision makers into groups, thus, allowing reducing the number of supported user models and increasing the data presentation quality. Finally, these results lead to increasing the decision quality, what is important when decisions have to be made under time pressure. To find the above inference the described here approach applies decision mining techniques as a subarea of data mining.
Analysis of different decisions’ kinds is one of the areas of data mining for business processes. The goal of decision mining is to find “rules” explaining under what circumstances certain activity is to be selected rather than the other one. For instance, decision mining aims at detecting data dependencies that affect the routing of a case in the event log of the business process executions (Rozinat and van der Aalst, 2006). There is a set of tools implementing different tasks of decision mining: “Decision Miner” (Rozinat and van der Aalst, 2006), Decision mining software “Risky Business” and “GoldPan” (Decision Mining software, 2007) and other.
Decision mining covers a wide range of problems. Estimation of data quality and interpretation of their semantics is one of the major tasks of decision mining. It requires the following interpretation of data: whether it is relevant, what it actually means, in what units it is measured, etc. Classification of decisions is also one of important tasks for decision mining. These tasks solving requires development of decision models and (semi)automatic decision analysis techniques. For instance, in the approach the concept of decision trees has been adapted to carry out a decision point analysis (Rozinat and van der Aalst, 2006), spatial analysis has been adapted for criminal event prediction (Brown et al., 2006).