|< Day Day Up >|| |
Guided knowledge discovery through interactive data mining as a discrete field of research is still in its infancy and as such there are few published works of relevance. This leaves open broad and diverse areas of further research in the areas of algorithmic development, interaction, and presentations and in associated areas such as collaborative guidance.
Future algorithmic research in the field will focus upon methods by which guidance can be incorporated into new or existing explorative algorithms at different levels of granularity. Preliminary areas that show promise include priority-based algorithms (Brin & Page, 1998), incremental computation (Sundaresh & Hudak, 1991) and state-based processing, which uses the concept of rollback to return to a previous intermediate state instead of re-instigating a new analysis. An associated area is the investigation of supporting frameworks that provide flexible interactive knowledge discovery environments (Roddick & Ceglar, 2001; Wrobel et al., 1996).
It seems likely that the majority of these techniques will remain domain-specific (if not task-specific) because of associated subjective interpretation. The challenge lies in the creation of generic sets of interaction mappings between the graphical interface and the underlying mining process. The development of such mappings was indicated by the research of the MERL team (D. Anderson, Anderson, Lesh, Marks, Perlin, Ratajczak & Ryall, 2000; Lesh, Marks & Patrignani, 2000) where a single set of interaction functions were effectively incorporated within two different problems domains, using different domain-specific presentations.
Existing knowledge discovery tools do not adequately provide the capabilities to incorporate subjective measures of interestingness into the analysis process. Current analysis results in ineffective discovery processes, as heuristic measures cannot accurately portray what is potentially of interest to the user. As shown by the MERL team (Anderson et al., 2000) and Brin and Page (1998), subjective judgement can be incorporated by actively engaging the user in the mining process. Benefits include an accelerated knowledge discovery process and improved results. User participation in the mining process results in greater confidence in the correctness of the discovered patterns due to the sense of control that guidance capabilities provide.
|< Day Day Up >|| |