thumnlab / dhpe

This is a sample implementation of "Dynamic High-order Proximity preserved Embedding (DHPE)" (TKDE 2018).

DHPE

This is a sample implementation of "Dynamic High-order Proximity preserved Embedding (DHPE)" (TKDE 2018).

Requirements

MATLAB

Usage

Static Model

Run embed_static.m with matlab as the static model of DHPE

Input:
    A: N*N adjacency matrix (sparse)
    K: dimensionality of embedding space
    beta: decaying constant, default is 0.8 / spectral radius
Output:
    U, S, V: the GSVD result of the high-order proximity (katz) matrix

The high-order proximity (katz) matrix is approximated by U * S * V' (see "Asymmetric Transitivity Preserving Graph Embedding", KDD 2016)

Dynamic Model

Run embed_update.m with matlab as the dynamic model of DHPE

Input:
    detA: N*N sparse matrix (the changed edges)
    U, S, V: the GSVD result of the high-order proximity (katz) matrix
    mA = Fa in paper 
    mB = Fb in paper
Output:
    nU, nS, nV: update the GSVD result of the high-order proximity (katz) matrix
    nA = update Fa in paper
    nB = update Fb in paper

Cite

If you find this code useful, please cite our paper:

@article{zhu2018high,
  title={High-order proximity preserved embedding for dynamic networks},
  author={Zhu, Dingyuan and Cui, Peng and Zhang, Ziwei and Pei, Jian and Zhu, Wenwu},
  journal={IEEE Transactions on Knowledge and Data Engineering},
  volume={30},
  number={11},
  pages={2134--2144},
  year={2018},
  publisher={IEEE}
}