----------------------------------------------------
RW ModMax
----------------------------------------------------
This package includes the implementation of the RW ModMax algorithm for features extraction on graphs.

 - RW_ModMax.m : compute the top eigenvectors of the random walk-based modularity matrix for use as features.
 - Zx.m : estimates the product of the fundamental bag-of-paths matrix with a column vector. Used by the RW_ModMax function.
 - example_run.m : example of the use of RW ModMax on a labelling problem on the BlogCatalog dataset.

The example run relies on the SocDim framework for semi-supervised classification on graphs developped by Lei Tang and Huan Liu.
The original SocDim framework is available at http://leitang.net/social_dimension.html
We include hereafter the description of the SocDim framework by Lei Tang.

Robin Devooght (feel free to contact me at [email protected] for more information)

------------------------------------------------------
SocDim Framework
------------------------------------------------------
This package includes the implementation of SocialDimension approach to classification with network data.

Related Reference:
@inproceedings{1557109,
 author = {Tang, Lei and Liu, Huan},
 title = {Relational learning via latent social dimensions},
 booktitle = {KDD '09: Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining},
 year = {2009},
 isbn = {978-1-60558-495-9},
 pages = {817--826},
 location = {Paris, France},
 doi = {http://doi.acm.org/10.1145/1557019.1557109},
 publisher = {ACM},
 address = {New York, NY, USA},
 }

sociodim.m: given the social dimensions, build a svm classifier and make predictions.

evaluate.m: evaluate the performance given prediction scores and the true label
In order to perform evaluation, you have to mex construct_indictor.c
An evaluation script testevaluation.m is provided to check whether the evaluation script works.

linearsvm.m: a wrapper code for liblinear
To bulid the classifier, liblinear package is required. 
http://www.csie.ntu.edu.tw/~cjlin/liblinear/
Please add the exectable file for training (train.mex) to the matlab path.
Current matlab code has been tested on liblinear 1.5. 
If you use an older version, please uncomment the code between lines 38-44 in linearsvm.m.

For convenience, the executable files (*.mexa64) for  linux64 are included in the package.