Overlapping Community Detection in Temporal Networks

Background/Objectives: One of the most commonly observed features of Online Social Networks is Community Structure. This feature provides great benefit focusing on insights of network structure, hidden patterns and the flow of information between actors. Methods/Statistical Analysis: Most real-world social networks are inherently dynamic, grow rapidly in terms of social interactions. These interactions in network are reflected by edges in a graph. Instead of updating a network structure for every edge change, the proposed method tracks the edges at every unique time stamp in a subgraph and modify the network only with the changed edges. Findings: There are many static community detection algorithms for discovering communities in networks, but very few deal with incremental structural changes in the network. The proposed algorithm DOMLPA (Dynamic Overlapping Multi-Label Propogation) deals with dynamic networks where data arrives as a stream to find the overlapping nodes in communities. To find the new edges the proposed algorithm lists out the differences in edges between the subgraph and the network for every snapshot. Based on the differences the label edges would be added or removed from the network and adjacency entries, neighbors’ list and label distribution entries are modified eventually. Speaker node function is activated to start the propogation process inorder to get the labels for every node. If a node contains a only one label it belongs to single community. If a node carries more than one label with more than one maximum probability entry, then it belongs to multiple communities. The opportunity to capture the evolutionary patterns in dynamic networks is lost by not considering the time of interaction in static algorithms. Application/Improvements: The study of communities is helpful in examining patterns leading to understand the structure of networks, finding the information flow and events taking place between a group of social actors over a period of time and to identify trending sentiments about brands based on tweets.