![Dataset
Railway dataset
Railway network, proposed by [Ghosh et al.
2011] consists of nodes representing railway
stations in India, where two stations si and sj
are connected by an edge if there exists at
least one train-route such that both si and sj
are scheduled halts on that route. Here the
communities are states/provinces of India since
the number of trains within each state is much
higher than the trains in-between two states.](https://image.slidesharecdn.com/ireproject23team24-160414095807/85/Integrating-Network-Discovery-and-Community-Detection-IRE-IIITH-Team-24-3-320.jpg)
![Dataset
Football
Football network, proposed by [Girvan and
Newman 2002a] contains the network of
American football games between Division IA
colleges during the regular season of Fall 2000.
The vertices in the graph represent teams
(identified by their college names) and edges
represent regular-season games between the
two teams they connect.](https://image.slidesharecdn.com/ireproject23team24-160414095807/85/Integrating-Network-Discovery-and-Community-Detection-IRE-IIITH-Team-24-4-320.jpg)
Integrating Network Discovery and Community Detection (IRE IIITH) Team 24
- 1. Team No 24 Integrating Network Discovery and Community Detection Nikhil Daliya - 201301142 Athresh G - 201505565
- 2. Overview Integrating network discovery and community detection routines for nodes in the given network and identifying the characteristics of the nodes (constant or rapidly changing) in the network.
- 3. Dataset Railway dataset Railway network, proposed by [Ghosh et al. 2011] consists of nodes representing railway stations in India, where two stations si and sj are connected by an edge if there exists at least one train-route such that both si and sj are scheduled halts on that route. Here the communities are states/provinces of India since the number of trains within each state is much higher than the trains in-between two states.
- 4. Dataset Football Football network, proposed by [Girvan and Newman 2002a] contains the network of American football games between Division IA colleges during the regular season of Fall 2000. The vertices in the graph represent teams (identified by their college names) and edges represent regular-season games between the two teams they connect.
- 5. Dataset Football The teams are divided into conferences (indicating communities) containing around 8- 12 teams each. Games are more frequent between members of the same conference than between members of different conferences. Teams that are geographically close to one another but belong to different conferences are more likely to play one another than teams separated by large geographic distances.
- 6. Application ●Exploring the adversarial networks(such as terrorist networks). ●Clustering in social networks. ●Politeness policies on crawling website makes it difficult to mine the whole network on social networking sites. There are space and bandwidth limits which put constraints on the size of network that can be mined.
- 7. Challenges ● Dynamic discovery of the network imposes problems in clustering of nodes . ● Identifying the characteristic of nodes(constant , changing or rapidly changing) is difficult problem. ●The dataset grows rapidly with network discovery and keeping track of probability distribution of each node for different communities is challenging task.
- 8. Tools Used ● Third party package (https://sites.google.com/site/santofortunato/in thepress2) for generating synthetic graphs as input. ● Language to be used: Python and Java. Packages such as panda, numpy, scikitlearn, networkX and igraph will be used accordingly. ● matplotlib package for plotting the results for better visualization and understanding.
- 9. Implementation ●We have used 2 modules mainly ChooseNode which chooses node in each iteration to be merged to the network and UpdateCommunity which will update the community or clusters from the choosen node. ●Spectral clustering is applied on the initial set of target nodes.
- 10. Implementation During ChooseNode we use 2 measures to choose the node for updation. Ncut measure : minimize the similarity across a cut, while simultaneously maximizing the similarity within the same community. Modularity : additional fraction of the edges that fall within the given communities over the expected fraction
- 11. Implementation I/P : ●Initial set of clustering , Initial network, cost and budget. O/P : ●Final network and nodes with clusters formed from nodes we have discovered. ●List of rapidly changing nodes in the network.
- 12. Results and Analysis - We have used Average Clustering Purity (ACP) and Average Clustering Entropy (ACE) to measure effectiveness of our algorithm. - Both these measures incorporates the fraction of nodes of particular cluster belonging to same class as their measure.
- 13. Results and Analysis Railway Dataset : Total no. of target nodes : 80 Average cluster purity : 0.79 Average Cluster entropy : 0.17 Rapidly changing nodes : 6,47,84,91
- 14. Results and Analysis Railway Dataset :
- 15. Results and Analysis Railway Dataset :
- 16. Results and Analysis Football Dataset : Total no. of target nodes : 48 Average cluster purity : 0.91 Average Cluster entropy : 0.11 Changing nodes : 51 , 63 , 49
- 17. Results and Analysis Football Dataset :
- 18. References Research paper : On integrating Network and Community Discovery http://hanj.cs.illinois.edu/pdf/wsdm15_jliu.pdf
- 19. Thank You !!!!