Abstract—This paper describes an interactive tool for constrained clustering that helps users to efficiently select effective constraints during the constrained clustering process. Constrained clustering is a promising technique for smart data aggregation or filtering, which is indispensable for the user activity on the Web. Effective bias is necessary for the constraints selection in order to make it a more practical technique, We approach this problem by incorporating human biasing using an easy manipulatable interactive tool. This tool has several functions such as the 2-D visual arrangement of a dataset and constraint assignment by mouse manipulation. Moreover, it can be used to execute distance metric learning and k-means clustering. In this paper, we show an overview of the tool and how it works, especially for the functions for display arrangement by using multi-dimensional scaling and incremental distance metric learning. In the experiments, we investigated the performance of the sampling heuristics found by observing the interaction between the users and our tool. The results show that the heuristic outperforms the random sampling method both in the two benchmark datasets from the UCI repository and a Web page dataset from the Open Directory Project.  

Cite: Masayuki Okabe and Seiji Yamada, "An Interactive Tool for Human Active Learning in Constrained Clustering," Journal of Emerging Technologies in Web Intelligence, Vol. 3, No. 1, pp. 20-27, February 2011. doi:10.4304/jetwi.3.1.20-27