Breed sorting and cemetery formation are two prominent examples of the complex tasks of clustering and sorting performed by colonies of ants in nature. Its mechanisms have inspired ant-based clustering, a meta-heuristics which has first been proposed by Deneuborg for the use in robotics. This first variant has, later on, been modified to extend to numerical data analysis and to graph clustering in particular. After a short introduction to the basic concepts of ant-based clustering and an outline of previous research, this talk will concentrate on the algorithm's application to the visualization of online Internet queries. In particular, we will study the dynamic generation of so called topic maps, a type of data visualization that has become increasingly popular and whose usability has only recently been demonstrated. Topic maps are based on the two-dimensional spatial representation of large data collections; making use of a landscape metaphor, semantic similarity is mapped to spatial proximity, and classifying labels are used as landmarks. The fast generation of such maps using ant-based clustering requires a number of fundamental modifications of the existing algorithms. Aside from the extensions necessary for the application to real document data, a significant speed-up is required to satisfy the time constraints posed by an online query-system. Several fundamental algorithmic changes introduced to meet these requirements are explained and discussed in detail. The talk will close with an analysis of the current restrictions of the approach and provide an outlook towards future work necessary to overcome these limitations.
Ant-based clustering, Data visualization, Topic maps, Multi-dimensional scaling