In the literature about association analysis, many interestingness measures have been proposed to assess the quality of obtained association rules in order to select a small set of the most interesting among them. In the particular case of hierarchically organized items and generalized association rules connecting them, a measure that dealt appropriately with the hierarchy would be advantageous. Here we present the further developments of a new class of such hierarchical interestingness measures and compare them with a large set of conventional measures and with three hierarchical pruning methods from the literature. The aim is to find interesting pairwise generalized association rules connecting the concepts of multiple ontologies. Interested in the broad empirical evaluation of interestingness measures, we compared the rules obtained by 37 methods on four real world data sets against predefined ground truth sets of associations. To this end, we adopted a framework of instance-based ontology matching and extended the set of performance measures by two novel measures: relation learning recall and precision which take into account hierarchical relationships.

This paper presents a data mining system for multi-label classification and hierarchy extraction from the predictions provided by a multi-label classifier. Classes in multi-label classification tasks are often hierarchically organized and the hierarchy is assumed to be known. A much less investigated approach and a more challenging task, however, is to suppose that the underlying class taxonomy is unknown and that a data mining system can infer it automatically. In our setting, the proposed system is trained with multi-label data and is subsequently able to produce multi-label predictions along with hierarchical relationships between classes. The hierarchy extraction algorithm is based on building association rules from label co-occurrences. Within the framework we examine the performance of two recently introduced multi-label extensions of Adaptive Resonance Theory (ART)-based neural networks: Multi-Label Fuzzy ARTMAP (ML-FAM) and Multi- Label Fuzzy Adaptive Resonance Associative Map (ML-ARAM) in comparison with two state-of-the-art classifiers Multi-Label k-Nearest Neighbors (ML-kNN) and BoosTexter, taking into account the quality of hierarchy extraction. We also develop a novel distance measure for the quantitative evaluation of the derived class hierarchies and compare it with two other distance measures. To demonstrate the effectiveness of the proposed approach, experiments on several benchmark datasets have been performed.

This paper presents a modified search procedure for the Adaptive Resonance Theory (ART) neural networks, that does not utilise the user-defined vigilance parameter. The original ART search is based on the iterative check of match criterion for the winning node against it. The winner is defined according to the winner-take-all activation as a node with the maximal value of the choice function. The search procedure usually requires several cycles until the satisfactory winning node is found. We propose to combine the choice function and the match criterion and choose a node that maximises both, so only one search cycle is necessary. In this case the search time can be reduced significantly and the network performs well enough without setting of vigilance.

Research into the application of distributed ARTMAP (dARTMAP) and fuzzy ARTMAP neural networks to quality testing is presented. The aim of this work is the classification of faulty wafers into one of 22 known classes based on topological information extracted from the geometric allocation of the defective chips on the wafer. The classification abilities of dARTMAP and fuzzy ARTMAP were investigated in comparison with the results for counterpropagation networks (CPN). Experimental results show that ART algorithms can be successfully applied to solve this problem. Both of the ART networks classify significantly better than CPN networks.