PROJECT TITLE :

Fuzzy Clustering in a Complex Network Based on Content Relevance and Link Structures

ABSTRACT:

Several real-world problems can be represented as complicated networks with nodes representing different objects and links between nodes representing relationships between objects. As completely different attributes will be thought-about as associating with different objects, alternative than nontrivial link structures, complicated networks conjointly contain made content information, and it will be a massive challenge to seek out interesting clusters in such networks by absolutely exploiting the knowledge of each content and link information in them. Although some attempts are created to tackle this clustering downside, few of them have thought of the feasibility of identifying clusters in complicated networks using a fuzzy-primarily based clustering approach. We believe that, if the degree of membership to a cluster that a node belongs to will be thought of, we can be ready to better establish clusters in complex networks, as we might be in a position to spot overlapping clusters. In this paper, we have a tendency to, so, propose a fuzzy-based mostly clustering algorithm for this task. The algorithm, that we call Fuzzy Clustering Algorithm for Complicated Networks (FCAN), can discover clusters by taking into consideration each link and content info. It will thus by 1st processing the content info by introducing a measure to quantify the relevance of contents between each combine of nodes among the network. It then proceeds to leverage the link data within the clustering process by considering a measure of cluster density. Primarily based on these measures, FCAN identifies fuzzy clusters that are additional densely connected and additional highly relevant in their contents to optimize the degrees of memberships of every node belonging to different clusters. The performance of FCAN has been evaluated with several synthetic and real datasets involving those of document classification and social community detection. The results show that, in terms of accuracy, computation potency, and scalability, FCAN will be a terribly promising approach.