作者: Dmitry I. Ignatov , Dmitry V. Gnatyshak , Sergei O. Kuznetsov , Boris G. Mirkin
DOI: 10.1007/S10994-015-5487-Y
关键词:
摘要: This paper presents several definitions of "optimal patterns" in triadic data and results experimental comparison five triclustering algorithms on real-world synthetic datasets. The evaluation is carried over such criteria as resource efficiency, noise tolerance quality scores involving cardinality, density, coverage, diversity the patterns. An ideal pattern a totally dense maximal cuboid (formal triconcept). Relaxations this notion under consideration are: OAC-triclusters; triclusters optimal with respect to least-square criterion; graph partitions obtained by using spectral clustering. We show that searching for an tricluster cover NP-complete problem, whereas determining number covers #P-complete. Our extensive computational experiments lead us clear strategy choosing solution at given dataset guided principle Pareto-optimality according proposed criteria.