Temporal-Structural Random Walk: Unifying Structural Proximity and Equivalence in Dynamic Networks

Installation

The code was tested on Python 3.7.11. The requirements are in setup.py.

cd temporal_structural_walk
pip install .

Usage

1. -p, --datapath (str, required):

   • Path to the input data file.

2. -s, --savedir (str, optional, default='.'):

   • Directory where results will be saved.

3. -d, --embedding_dimension (int, optional, default=32):

   • Embedding dimensions.

4. -r, --random_walks_per_node (int, optional, default=20):

   • Number of random walks per node.

5. -l, --maximum_walk_length (int, optional, default=20):

   • Maximum length of eac random walk.

6. -w, --context_window_size (int, optional, default=10):

   • Size of the context window for generating embeddings.

7. -k, --k (int, optional, default=-1):

   • Top k neighbors to keep in structural similarity network.

8. -z, --save_embeddings (int, optional, default=0):

   • Flag to save embeddings (0 = no only show the evaluation scores, 1 = yes).

9. --class_type (str, optional, default='multiclass'):

   • Type of classification problem (multiclass, binary, multilabel).

10. --alpha (float, optional, default=0.0):

    • Balancing weight in range [0, 1]. When set to 0, the algorithm only considers proximity and the random walk performed is reduced to a temporal walk. When set to 1, the algorithm only considers equivalence, the random walk performed is equivalent to a random walk on a structural similarity network S.

To reproduce the result for Enron dataset, run

python src/temporal_structural_walk/run_embedding.py -p data/hospital -d 32 -r 20 -l 25 -w 10 -k 5 --alpha 0.05 --class_type multiclass

To reproduce the result for Workplace dataset, run

python src/temporal_structural_walk/run_embedding.py -p data/hospital -d 32 -r 20 -l 15 -w 10 -k 5 --alpha 0.025 --class_type multiclass

To reproduce the result for Enron dataset, run

python src/temporal_structural_walk/run_embedding.py -p data/enron -d 32 -r 20 -l 20 -w 10 -k 5 --alpha 0.1 --class_type multiclass

To reproduce the result for PPI-aging dataset, run

python src/temporal_structural_walk/run_embedding.py -p data/aging -d 32 -r 20 -l 30 -w 10 -k 100 --alpha 0.95 --class_type binary

To reproduce the result for Brain dataset, run

python src/temporal_structural_walk/run_embedding.py -p data/brain -d 32 -r 20 -l 10 -w 10 -k 20 --alpha 0.05 --class_type multiclass