Lehr- und Forschungseinheit für Datenbanksysteme Ludwig-Maximilians-Universität München
Institut für Informatik
Lehr- und Forschungseinheit für Datenbanksysteme
University of Munich
Institute for Computer Science
Database and Information Systems

Visual Data Mining


One popular definition of Knowledge Discovery in Databases is the non-trivial process of identifying valid, novel, potentially useful, and understandable patterns in data. Data mining which can be seen as the core of the KDD process actually maps the data to some kind of valid, novel, potentially useful and understandable knowledge. Obviously, just the user can determine whether the resulting knowledge satisfies these requirements. Moreover, the usefulness of some kind of knowledge varies from user to user. The new area of visual data mining focuses on integrating the user in the KDD process in terms of effective and efficient visualization techniques, interaction capabilities and knowledge transfer.


There are different approaches for integrating the user in the KDD process. Thus we can classify our projects in the way of their role in DM:

Fully Automatic Data Mining (FADM)

In most existing cases, DM is done completely algorithmic. Our research group explores different algorithmic approaches to similarity search for the sake of appropriate feedback for the user. Typically, similarity search is an iterative process and the visualization task is to find an intuitive representation of the similarity model such that it provides an explanation component for the retrieved objects as well as a query component based on the refinement of the last query. 

Semi-Automatic Data Mining (SADM)

This project divides DM into two parts, one of which is performed algorithmic, the other one is interactive. Chaining up both parts can yield a large gain in effectivity and efficiency in comparison to completely algorithmic approaches. With OPTICS, a new approach to cluster analysis has been developed which identifies the density-based clustering structure automatically. The ordering of the objects obtained by the algorithm is used for the visualization as an appropriate arrangement of the objects.

Interactive Data Mining (IDM)

Our PBC system (Perception-Based Classification) maps the problem of finding an accurate decision tree on a visualization technique with interaction capabilities. This approach entails several advantages over existing algorithmic approaches. PBC enables backtracking during the tree construction and an arbitrary amount of split points. Since the decision tree is constructed by the user utilizing his perception, (domain) knowledge is transfered in both directions.
For access to the Web-based version of PBC, please contact Mihael Ankerst.


List of Papers