A clustering-based adaptive Neighborhood Retrieval Visualizer. (August 2021)

• Record Type: Journal Article
• Authors: Olszewski, Dominik
• Is Part Of: Neural networks. Volume 140(2021)
• Page Start: 247
• Page End: 260
• Publication Date: 2021-08
• Subjects: Neighborhood Retrieval Visualizer -- Adaptive Neighborhood Retrieval Visualizer -- Information retrieval -- Data clustering -- Data visualization; Neural computers -- Periodicals; Neural networks (Computer science) -- Periodicals; Neural networks (Neurobiology) -- Periodicals; Nervous System -- Periodicals; Ordinateurs neuronaux -- Périodiques; Réseaux neuronaux (Informatique) -- Périodiques; Réseaux neuronaux (Neurobiologie) -- Périodiques; Neural computers; Neural networks (Computer science); Neural networks (Neurobiology); Periodicals; 006.32
• DOI: 10.1016/j.neunet.2021.03.018
• Languages: English
• ISSNs: 0893-6080

Abstract

Abstract: We introduce a novel adaptive version of the Neighborhood Retrieval Visualizer (NeRV). We maintain the advantages of the conventional NeRV method, while proposing an improvement of the data samples' neighborhood width calculation, in the input and output data space. In the standard NeRV, the data samples' neighborhood widths are determined in an arbitrary manner, in this way, inhibiting the possible quality of the resulting data visualization. We propose to compute the widths adaptively, on the basis of the input data scattering. Therefore, we first perform the preliminary input data clustering, next, we calculate the values of the inner-cluster variances, which convey the information on the input data scattering, then, we assign them to each data sample, and finally, we use them as the basis for the data samples' neighborhood widths determination. The results of the experiments conducted on the three different real datasets confirm the effectiveness and usefulness of the proposed approach. Highlights: An improved adaptive version of the Neighborhood Retrieval Visualizer is proposed. The introduced visualization takes into account the input data scattering and adjusts to it correctly. Data samples' neighborhood widths are determined on the basis of the inner-cluster variance of the data in the input space. The inner-cluster variance is calculated after the preliminary data clustering carried out in the input data space. The experimental research conducted on three real datasets confirmed the usefulness and effectiveness of the proposed approach.