Publication Date

Spring 2021

Degree Type

Master's Project

Degree Name

Master of Science in Computer Science (MSCS)


Computer Science

First Advisor

Katerina Potika

Second Advisor

Christopher Pollett

Third Advisor

William Andreopoulos


community intersection, social networks


Social networking sites are important to connect with the world virtually. As the number of users accessing these sites increase, the data and information keeps on increasing. There are communities and groups which are formed virtually based on different factors. We can visualize these communities as networks of users or nodes and the relationships or connections between them as edges. This helps in evaluating and analyzing different factors that influence community formation in such a dense network. Community detection helps in revealing certain characteristics which makes these groups in the network unique and different from one another. We can use such information to find trends in the network which might help in understanding complex systems.

In this project, we will study the problem of detecting local overlapping commu- nities in a stream graph, by proposing a new metric that of common communities of the endpoints of a new edge. Moreover, we discover good seed nodes by finding an offline non overlapping community structure of a small sub graph as a preprocess- ing step. Additionally, we also evaluate these methods with different and extensive datasets. We experiment with a new web graph dataset along with some other more commonly used datasets. F1 score is used for datasets that have the ground truth. The proposed algorithm outperforms the traditional methods by 17%. For the new web graph dataset we use the overlapping modularity metric to evaluate our approach. This approach yields accurate modularity scores up to 0.71 for the web graph dataset increasing the accuracy by 10%. In the end, we discuss the approaches used and their results along with scope to improve for the future.