Experiences in building a tool for navigating association rule result sets
MetadataShow full item record
Practical knowledge discovery is an iterative process. First, the experiences gained from one mining run are used to inform the parameter setting and the dataset and attribute selection for subsequent runs. Second, additional data, either incremental additions to existing datasets or the inclusion of additional attributes means that the mining process is reinvoked, perhaps numerous times. Reducing the number of iterations, improving the accuracy of parameter setting and making the results of the mining run more clearly understandable can thus significantly speed up the discovery process. In this paper we discuss our experiences in this area and present a system that helps the user to navigate through association rule result sets in a way that makes it easier to find useful results from a large result set. We present several techniques that experience has shown us to be useful. The prototype system – IRSetNav – is discussed, which has capabilities in redundant rule reduction, subjective interestingness evaluation, item and itemset pruning, related information searching, text-based itemset and rule visualisation, hierarchy based searching and tracking changes between data sets using a knowledge base. Techniques also discussed in the paper, but not yet accommodated into IRSetNav, include input schema selection, longitudinal ruleset analysis and graphical visualisation techniques.