MMDW

Max-margin DeepWalk

Code of IJCAI2016: "Max-Margin DeepWalk: Discriminative Learning of Network Representation"

Datasets

We evaluate MMDW with three datasets, including Cora, Citeseer and Wiki.

• data/Edgelist/*_edgelist.txt: original edgelist file of network *.
• data/Category/*_category.txt: the category list of vertices.
• data/Net/*_net.txt: the matrix M of obtained from transition matrix A of a network.
• data/vector/: the folder to save learnt vectors of vertices.
• data/svm_model/: the folder to save trained svm classifiers.
• data/Bias/: the folder to save calculated biasVectors.
• data/result/: the folder to classification results.

Run

Run the following command to learn max-margin DeepWalk:

java -jar mmdw.jar "dataset" "data_folder" "order_of_alphaBias" 

Here is an example:

java -jar mmdw.jar Cora data/ -3 

Cite

If you use the code, please cite this paper:

Cunchao Tu, Weicheng Zhang, Zhiyuan Liu, Maosong Sun. Max-Margin DeepWalk: Discriminative Learning of Network Representation. International Joint Conference on Artificial Intelligence (IJCAI 2016).

For more related works on network representation learning, please refer to my homepage.