A Temporal User Attribute-Based Algorithm to Detect Communities in Online Social Networks

Abstract

The world is witnessing the daily emergence of a vast variety of online social networks and community detection problem is a major research area in online social network studies. The existing community detection algorithms are mostly edge-based and are evaluated using the modularity metric benchmarks. However, these algorithms have two inherent limitations. Firstly, they are based on a pure mathematical object which considers the number of connections in each community as the main measures. Consequently, a resolution limit and low accuracy in finding community members in often observed. Whereas, online social networks are dynamic networks and the key players are humans whose main attributes such as lifespan, geo-location, the density of interactions, and user weight, change over time. These attributes tend to influence the formation of user communities in any category of online social network. Secondly, the output structure of existing community detection algorithms is usually provided as a graph and dendrogram. A graph structure, is, however, characterized by a high memory complexity, and subsequently exponential search time complexity. Implementing dendrogram such a complex structure is complicated. To address memory complexity and the accuracy rate of the community detection issues, this paper proposes a new temporal user attribute-based algorithm, namely the recently largest interaction based on the attributes of a typical online social network user. Experimental results show that the proposed algorithm outperforms eight well-known algorithms in this domain.

Strategies for improving the quality of community detection based on modularity optimization

“…In literature, Louvain's algorithm is the most robust modularity optimization algorithm, so it is often referred to for further research [20], [70]- [73]. Forster [70] developed the Louvain algorithm to support parallel computing.…”

“…The first case is called solution degeneration, where a large number of community structures may be found from a network whose topological structures are very different from one another but produce a value of modularity that is very close to the optimal [18], [19]. The second case is called the resolution limit, which is a limitation in finding communities that are smaller than a certain scale [20].…”

Connecting User Profiles of Social Networks Using Proximity-Based Clustering

“…A community network was formed based on the user attributes such as location, high-density interactions between users, user lifespan, and user weight. The proposed method Recently largest Information [36] had better performance in identifying the network structure for time complexity and accuracy based on identified user attributes over the existing methods in a dynamic social network.…”