IDEAS Lab UT's submission to the WSDM 2022 Temporal Link Prediction challenge.
To generate our predictions, run the following two Jupyter notebooks:
construct_features.ipynb
: This will construct the edge features for the final test set and save them to CSV.predict_edges.ipynb
: This will train a logistic regression model on the edge features and generate predictions on the final test set.
The feature construction may take a few hours, so we have included also the constructed features with the filenames beginning with featureA_
and featureB_
so you can directly run predict_edges.ipynb
to generate predictions from our pre-computed features.
The following input data files are assumed to be in the root directory:
edges_train_A.csv
node_features.csv
edge_type_features.csv
input_A_initial.csv
input_A.csv
edges_train_B.csv
input_B_initial.csv
input_B.csv
Predictions will be generated also in the root directory:
output_A.csv
output_B.csv
See requirements.txt
for required packages. The code for the CHIP model is included in the directory CHIP-Network-Model
, which is added to the system path when loading chip_features.py
.