Knowledge refreshing: Model, heuristics and applications

Abstract

With the wide application of information technology in organizations, especially the rapid growth of E-Business, masses of data have been accumulated. Knowledge Discovery in Databases (KDD) gives organizations the tools to sift through vast data stores to extract knowledge supporting organizational decision making. Most of the KDD research has assumed that data is static and focused on either efficiency improvement of the KDD process (e.g., designing more efficient KDD algorithms) or business applications of KDD. However, data is dynamic in reality (i.e., new data continuously added in). Knowledge discovered using KDD becomes obsolete rapidly, as the discovered knowledge only reflects the status of its dynamic data source when running KDD. Newly added data could bring in new knowledge or invalidate some discovered knowledge. To support effective decision making, knowledge discovered using KDD needs to be updated along with its dynamic data source. In this dissertation, we research on knowledge refreshing, which we define as the process to keep knowledge discovered using KDD up-to-date with its dynamic data source. We propose an analytical model based on the theory of Markov decision process, solutions and heuristics for the knowledge refreshing problem. We also research on how to apply KDD to such application areas as intelligent web portal design and network content management. The knowledge refreshing research identifies and solves a fundamental and general problem appearing in all KDD applications; while the applied KDD research provides a test environment for solutions resulted from the knowledge refreshing research.