We have a social network graph that represents citation network. Each node contains information about the author’s paper, title and topic, …,etc. If an author references another paper, an edge is existed.
My question is, could we classify this kind of graphs with normal classification algorithm ( i.e. Bayesian Classification, Decision Tree Induction, k-Nearest-Neighbor Classifiers, SVM, …, etc) or should we add more information to the graph.
It depends on what you wanna do with the graph. Do you want to classify subgraphs? Then I would recommend to use a similarity measure, e.g. a graph kernel, which enables to apply standard classification algorithms. I would need more details to help you out more specifically.
thanks Mr. Moeves. let's say we want to classify the network based on the category of the work. assume that each work belongs to specific category such as data mining, sensor network, distributed system, mobile computing, pervasive computing ... and so on. These information were implicity given and may be found in title or topic of the work.
So could we extract these information from the graph and then classify the social network.
you can add your assumptions as well.
So, as I understand you deal with many different networks where each one is labeled by a category. Now you want to learn a classification model. Then you can still consider graph kernels: http://jmlr.csail.mit.edu/papers/v11/vishwanathan10a.html
Alternatively you could drescribe your graph by graph measures, e.g. density, number of components, etc.
Every distinct label can be represented, e.g. by an integer. Then you could use any kind of classifier that works with kernels, e.g. an SVM. Is that clear to you?
Hi, this is a very interesting question.
Would you share your work on this problem?
Thanks.
Join us in Granada at SAMSIN 2019
http://emergingtechnet.org/SAMSN2019/#
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification