Loading...

A Comprehensive Method for Clustering Evolutionary Big Graphs

Yazdani Jahromi, Mehdi | 2020

737 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 53167 (01)
  4. University: Sharif University of Technology
  5. Department: Industrial Engineering
  6. Advisor(s): Khedmati, Majid
  7. Abstract:
  8. Today, many real-world datasets such as social network data and web pages can be shown as graphs. Community detection and clustering of these big graphs has many applications in different fields like recommender systems in social networks and Diag- nosis of diseases in communication networks among proteins. A cluster in a graph is a sub-graph with many internal and few external edges. A new method for local cluster detection around an existing vertex is introduced in this paper. This method applies random walk algorithm for cluster detection. The time complexity of this algorithm based on the graph size is polynomial. Therefore, it can be used for clustering of big graphs. The experimental evaluations on real world and synthetic graphs have shown that this algorithm has high performance for detecting communities with high density. A cluster ideally is a complete subgraph in graphs. However, in many applications including social networks this notion is overly restrictive due to requirement of existing all pairwise edges between nodes in a certain cluster. In this paper a novel approach based on stopping problem and ran- dom walks on graphs for local graph clustering is proposed. In this approach random walk agent is walking on the graph until certain condition based on dynamic programming stopping problem is met, then nodes which has been visited in the random walk process will be considered as initial cluster of the starting node. This approach results in significantly better performance in graph clustering compared to other algorithms in this area. At the end, the applicability of the proposed method is illustrated by real world examples
  9. Keywords:
  10. Random Walk ; Social Networks ; Graph Clustering ; Graph Partitioning ; Local Graph Clustering ; Stopping Problem ; Clique Detection

 Digital Object List

 Bookmark

...see more