Multi-criteria-based fusion for clustering texts and images case study on Flickr

Abstract

PurposeThis study aims to evaluate a new fusion technique of visual and textual clusters of objects from a real multimedia data-driven collection to improve the performance of multimedia applications.Design/methodology/approachThe authors focused on using multi-criteria for clustering texts and images. The algorithm consists of these steps: first is text representation using the statistical method of weighting, second is image representation using a bag of words feature descriptors methods and finally application of multi-criteria clustering.FindingsAs an application for event detection based on social multimedia data, in particular, Flickr platform. Several experiments were conducted to choose the appropriate parameters for a better scheme of clustering. The new approach achieves better performance when aggregate text clustering is done with image clustering for event detection.Research limitations/implicationsFurther researches would be investigated on other social media platforms such as Facebook and Twitter for a generalization of the technique.Originality/valueThis study contributes to multimedia data mining through the new fusion technique of clustering. The technique has its root in such strong field as the field of multi-criteria clustering and decision-making support.