Abstract

Community identification is an important technique for the investigation of complex networks because it makes it possible to examine mesoscopic features that are often connected to the organisational and functional properties of the underlying networks. In social media, there are billions of vertices and a variety of connections, thus community identification is a widely acknowledged approach of addressing the problem of grouping users. However, traditional methods are insufficient because of this. For the purpose of analysing social networks, overlapping community identification is crucial. On networks with complicated weight distributions, the current overlapping community recognition techniques seldom provide good results. Communities of any form may be easily and precisely found using density peaks clustering (DPC). Nevertheless, it also uses the truncation distance, and therefore is unable to automatically determine where the cluster centre is located. In this research work, an Adaptive density peak clustering (ADPC) with Modified Dragonfly Optimization (MDO) a suggested algorithm will decide the communities in a social network in an adaptable manner. Initially, the preprocessing methods such as Stemming, Stop-words removal, and Tokenization by bigrams, 1-to-3 grammars, and the unigram are three distinct types of data formats. Two feature extraction filters: Word Embedding Feature Extraction, and Term Frequency-Inverse Document Frequency (TF-IDF). In the Adaptive density peak clustering (ADPC) with Modified Dragonfly Optimization (MDO) algorithm, the clustering process is completed by determining the clustering centre by the MDO after determining the new local density based on the new neighborhood connection. ADPC-MDO adds a unique distance function based on common nodes to estimate the distance between nodes and takes weights into account to handle both weighted and unweighted social networks. A technique based on transitive consensus matrix building is used to generate a consensus matrix, which provides representative information of all dendrograms. Given a social networks dataset with a complicated weight distribution, the results show that the suggested ADPC-MDO performs better.