Document clustering is the process of partitioning a set of unlabeled documents into clusters such that documents within each cluster share some common concepts. To help with this analysis, concepts are conveniently represented using some key terms. For clustering algorithm, the most costly CPU time has to do with the classification phase. Using words as features, text data are represented in a very high dimensional vector space. We have studied a comparative advantage based algorithm for clustering sparse data in this space, it used one ¿ruler¿ instead of k centers to identify the comparative advantage of each cluster and define the cluster label for each document. However, this algorithm only considered the relative strength between clusters, the relationship between terms was ignored. In this paper, we proposed a weighted comparative advantage based clustering algorithm. The experimental results based on SMART system databases show that the new algorithm is better than simple comparative advantage algorithm, without any extra computation time. Compare with k-means, not only can it get comparable results but it can also significantly accelerate the clustering procedure.

MOTIVATION: Almost all methods of chromosome recognition assume supervised training; i.e. we are given correctly classified chromosomes to start the training phase. Noise, if any, is confined only in the representation of the chromosomes and not in the classification of the chromosomes. During the recognition phase, the problem is simply to calculate the string edit distance of the unknowns to the representatives chosen from the training phase and classify the unknowns accordingly. RESULTS: In this paper, a general method to tackle the difficult unsupervised induction problem is described. The success of the method is demonstrated by showing how the inductive agent learns weights in a dynamic manner that allows it to distinguish between noisy median and telocentric chromosomes without knowing their proper labels. The process of learning is characterized as the process of finding the right distance function, i.e. the distance function that can nicely separate the classes.

A general (nonheuristic) computational analytical model to tackle the difficult unsupervised inductive learning problem is proposed by making some additions and modifications to an existing metric model so that the model is more elegant and able to handle the unsupervised case. It turns out that it is instructive to treat, in essence, the supervised problem with noise as an unsupervised problem. We demonstrate the success of the new model on the benchmark XOR (exclusive-or) and parity problems by showing how the inductive agent successfully learns the weights in a dynamic manner that would allow it to distinguish between bit-strings of any length and unknown labels.