NIKE-ADIDAS / Signed-Graph-Contrastive-Learning

Signed-Graph-Contrastive-Learning

• Link to the paper - https://dl.acm.org/doi/pdf/10.1145/3459637.3482478

Data

• This file contains the implemetation of Signed Graph Contrastive learning on Bitcoin_alpha dataset available on https://snap.stanford.edu/data/soc-sign-bitcoin-alpha.html

• bitcoin_alpha\soc-sign-bitcoinalpha.csv - contains the original data converted to csv format conatining information about the nodes and labels.

• bitcoin_alpha\user_dict.pkl - contains dicionary of all the unique nodes present in the data.

• bitcoin_alpha\g_train.pkl - contains the dictionary of following format -: "items (('user', 'positive', 'user'), (array([ 0, 1, 2, ..., 7800, 1854, 10135]), array([ 483, 483, 483, ..., 10192, 10192, 4365]))) items (('user', 'negative', 'user'), (array([ 119, 120, 121, ..., 2727, 10059, 7112]), array([ 483, 483, 483, ..., 9547, 9496, 1854])))" here positive and negative indicate the sign of edges and users denotes the nodes.

• bitcoin_alpha\label_train.pkl - it contains the csv file on which training has to be done .

Code

• IPYNB file - sgcl_bitcoin_alpha.ipynb
• this code can run for other datasets as well mentioned in the repo by just changing the folder name in the code . Ex- sgcl_wikirfa.ipynb

Subparts in the code

• Importing necessary libraries
• Load user and features
• Load Train Graph
• Load Labels
• Model definition
• Graph Augmentation
• Define Predictor
• Training Parameter Setting
• Training
• Results from SGCL

Training on Logistic Regression, SVM and RandomForest on embeddings using various functions.

• i.HADAMARD
• ii. Concatenated
• iii. L1-normalization
• iv. L2-normalization
• v. Averages