Please use this identifier to cite or link to this item:
|Title:||Game theory based algorithms for community detection||Authors:||Radhika Arava||Keywords:||DRNTU::Science::Mathematics::Discrete mathematics::Algorithms||Issue Date:||2015||Source:||Radhika Arava. (2015). Game theory based algorithms for community detection. Doctoral thesis, Nanyang Technological University, Singapore.||Abstract:||The problem of community detection is important as it helps in understanding the spread of information in a social network. Linkages are more likely to form between similar people, leading to the formation of some community structure which characterizes the network dynamic. The more friends the two people have in common, the more the influence that each person can exercise on the other. We assume that communities capture homophily as people of the same community share a lot of similar features and hence the people of the same community are likely to follow the same trend. We use the concept of weighted potential games to formulate the model and the community detection algorithms. We propose a disjoint community detection algorithm, NashDisjoint that detects disjoint communities in any given network, which works as good as the state of the art algorithms on LFR Benchmarks for the mixing factors less than 0.6. We propose an overlapping community detection algorithm NashOverlap to detect the overlapping communities in any given network. We evaluate the algorithm NashOverlap against the state of the art algorithms so far and we find that our algorithm works far better than the state of the art on the standard LFR benchmarks in around 152 different scenarios, generated by varying the number of vertices, community size, mixing factor and overlapping membership with respect to the Normalized Mutual Information measure. We identify and study the significant collaboration groups of DBLP datasets using our algorithm NashOverlap. We compare our results with that of the algorithm COPRA and find that our algorithm can give more and better insights into the dataset.||URI:||https://hdl.handle.net/10356/65384||DOI:||10.32657/10356/65384||Fulltext Permission:||open||Fulltext Availability:||With Fulltext|
|Appears in Collections:||SPMS Theses|
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.