Summarizing Association Patterns Efficiently by Using PI Tree in a Data Stream Environment

Abstract

In the age of Knowledge economy, people are paying more attention to data mining; especially association patterns mining in owing to valuable information have often been buried with irrelevant noise. However, the number of the association patterns often exceeds the capacity of human's mind. Therefore, it is necessary to present patterns according to their interestingness effectively. On the other hand, the interestingness of patterns vary from different situation; people are interesting in revealing undiscovered truths, but also boring when they are familiar with them. Therefore, this approach focuses on continuously differentiating interesting and valuable patterns from data stream, which is the issue for Interestingness-oriented Knowledge Discovery (IOKD). Several principles and interestingness measures are proposed for modeling and measuring the interestingness of association patterns in a data stream. A new data structure, Pattern's Interestingness Tree (PI-Tree) is presented for discovering frequent patterns and helping to distinguish interesting knowledge. Performance Analysis indicates that the proposed approach is efficient for IOKD.