Monthly Archive for July, 2005

R. T. Ng and J. Han. Efficient and effective clustering methods for spatial data mining. In Proceedings of the 1994 International Conference Very Large Data Bases (Sept. 12-15, Santiago, Chile), pages 144–155. Morgan Kaufmann, San Francisco, CA, 1994. [url]

——————–

This article describes the CLARANS algorithm that is used to cluster spatial databases and that is based on randomised search. Problems of most of the methods used so far is that they require an a priori knowledge to be initialised. Their approach, on the contrary starts from scratch.

The method was developed from the CLARA algorithm, developed by Kaufmann et. al. The authors expand from this adding suppost for randomised search. Initially their method is able to find the best k_nat, which is the most natural number of clusters. Subsequently, the mothod start assigning objects to the clusters and finding the medoids.

Two versions of the method, respectively for Spatial and Non-Spatial-Dominant are tested against CLARA in a real estate data set. The results confirm the efficacy of the new method.

Tags: clustering, spatial clustering

K. Kopersky, J. Han, and J. Adhikary. Mining knowledge in geographical data. Communications of the ACM, pages 1–8, 1998. [url]

———————–

Data mining represent the confluence of several research fields, as machine learning, database systems, data visualization, statistics, and information theory.

The authors define the spatial data mining as the extraction of implicit knowledge, spatial relationships, or other patterns not explicitly stored in spatial databases. These data distinguish from the relational databases because they carry topological information, usually organised by a multidimensional spatial indexing structures.

The authors introduce several methods for knowledge mining in geographical data: the first is the generalization-based mining, the attempt to generalise abstract data from a low concept level. This method can be implemented with two phylosopies, a spatial-data-dominant generalization and a non-spatial-dominant generalization, which differ on the importance that is given to the spatial dimension of the data.

Another methodology introduced is the clustering, or densely populated regions, according to some distance measurement, in a large multi-dimensional data set.

A third methodology is the exploration of spatial associations, the rules that associate one or more spatial objects with other spatial objects. Some threshold can be implemented to control the filtering out of associations of objects.

Tags: spatial clustering

T. Honkela, S. Kaski, K. Lagus, and T. Kohonen. Newsgroup exploration with websom method and browsing interface. Technical Report A32, Helsinki University of Technology, Laboratory of Computer and Information Science, Rakentajanaukio 2 C, SF-02150 Espoo, Finland, 1996. [url]

———————-

This paper present the application of the SOM method to the exploration of a large number of usenet messages. This method, introduced by Kohonen, in 1982 is a means for automatically arranging high-dimensional statistical data so that alike inputs are in general mapped close to each other.

The general concept is that words are first organised into categories on a word category map, then an encoding of the documents can be achieved that explicity expresses the similarity of the word meanings.

This word category map os a self-organising semantic map, in the definition of Ritter and Kohonen, 1989, that describes relations of words based on their averaged short contexts. The SOM is a supervised and calibrated method. The document map is then formed with the SOM algorithm using the histograms as fingerprints of the documents.

Copyright notice: the present content was taken from the following URL, the copyrights are reserved by the respective author/s.

Tags: clustering, map algorithms, maps