Polysemous-Network-Embedding

This project provides an implementation for the paper:

Is a Single Vector Enough? Exploring Node Polysemy for Network Embedding
Ninghao Liu, Qiaoyu Tan, Yuening Li, Hongxia Yang, Jingren Zhou, Xia Hu
KDD 2019

Current implementation contains of two models:
PolyDeepwalk: Extends the Deepwalk model to handle different node aspects for homogeneous networks.
PolyPTE: Extends the PTE model to handle different node aspects for hetergeneous networks (bipartite networks in this work).

Files in the folder

• data/
  • BlogCatelog/
    • training.mat: training data samples in the form of (row, col, 1);
    • testing.pkl: testing data in python dict format, where dict[row]=[col1, col2, ...];
    • snmf6.mat: pre-obtained symmetric NMF clustering result on training data graph (this example has 6 clusters), following the model in https://github.com/dakuang/symnmf;
  • movielens/
    • training.mat: training data samples in the form of (row, col, val);
    • testing.mat: testing data samples;
    • sparsemat.npz: the sparse matrix verion of the training data;
• src/: implementations of polysemous embedding models.