Comparison of OA-biclustering algorithms

<p>This paper continues a series of studies on biclustering based on concept lattices. Biclustering is one of the key research tools in the field of data analysis. In view of the fact that many problems require the maintenance of the OA-description of cluster similarity, biclustering methods are widely used in many fields of science.</p><p>These methods are effective in solving the following problems: text mining and information retrieval, analysis of gene expression data, exploring communities, recommendation systems, analysis of site traffic, etc. The advantage of biclustering methods is that they allow to analyze the obtained clusters in respect to common attributes of the objects belonging to the same cluster.</p><p>The purpose of this paper is to conduct a comparative analysis of four biclustering algorithms: Ganter algorithm which computes all formal concepts of the given context; algorithm MyBiclusters, based on the object and attribute closures; recursive method of spectral clustering and the original algorithm Triage, developed by the author. The comparison is carried out in terms of average density, coverage, diversity and the number of generated biclusters. The relevance of the study lies in the novelty of such comparison of these algorithms and in the practical benefit of the developed algorithm.</p><p>Even though it is difficult to distinguish an optimal method among the algorithms under comparison, our computational experiments clearly show that the two methods based on object and attribute closures (MyBiclusters and Triage) are superior to the other. This fact tells of the high performance of these algorithms.</p>