Investigating the potential of art neural network models for indexing and information retrieval

Database management systems are very sophisticated, efficient, and fast in information retrieval tasks involving traditional data sets such as numbers, strings, and so on, but many limitations become evident when the data are more complex, that is, high or nondimensional data. Considering some existing problems in information retrieval processes, this work proposes a hybrid system that combines a model of the ART family neural network, ART2-A, with the Slim-Tree data structure, which is a metric access method. This approach is an alternative to perform clustering on data in an intelligent way so that the data can be recovered from the corresponding Slim-Tree. The proposed hybrid system is able to perform range and k-nearest neighbor queries, which is not an inherent characteristic in implementations involving artificial neural networks. Furthermore, experimental results showed that the performance of the hybrid system was better than the performance of Slim-Tree. © 2007 Wiley Periodicals, Inc. Int J Int Syst 22: 319–336, 2007.