FOCUSED INFORMATION CRITERION BASED PARTITIONED ITERATIVE X-MEANS DICE CORRELATION CLUSTERING FOR BIG GEO-SOCIAL DATA

Abstract

Geo-social data is location-based social media data which is generated by people on social network (i.e. face book, twitter etc.,) that is related to specific locations. There are lot of social users who are generates very large amount of data called “Big Data” that is difficult to be analyzed and make real-time decisions. Few research works have been designed for clustering geo-social data using different techniques. However, clustering performance of conventional algorithms was not higher to exactly find frequently visited location of users in social network when taking big geo-social dataset as input. In order to overcome such drawbacks, a Focused Information Criterion based Partitioned Iterative X-means Dice Correlation Data Clustering (FIC-PIXDCDC) Method is proposed in this work. The FIC-PIXDCDC Method groups the similar geo-social data with higher accuracy and lesser time. In FIC-PIXDCDC method, geo-social data (i.e., user, location and time) from Weeplaces dataset is initially taken as an input. After obtaining input, FIC-PIXDCDC method chooses number of clusters and centroids randomly. Then, FIC-PIXDCDC calculates dice correlation between each input geo-social data and cluster centroids. Subsequently, FIC-PIXDCDC method applies Focused Information Criterion to construct optimal number of clusters for a given big dataset. This process of FIC-PIXDCDC method is repetitive until no deviation in cluster centroids. Accordingly, FIC-PIXDCDC method group’s interrelated geo-social data together with higher accuracy and lower time to precisely discover location information of frequently visited users in social network. Experimental evaluation of FIC-PIXDCDC method is carried out on factors such as clustering time, clustering accuracy, error rate with respect to number of geo-social data.