This repository contains the instructions and materials necessary for reproducing the experiments presented in the paper: Benchmarking Network Embedding Models for Link Prediction: Are We Making Progress?
The repository is maintained by Alexandru Mara ([email protected]).
In order to reproduce the experiments presented in the paper the following steps are necessary:
-
Download and install the EvalNE library v0.3.2 as instructed by the authors here
-
Download and install the implementations of the baseline methods reported in the manuscript. We recommend that each method is installed in a unique virtual environment to ensure that the right dependencies are used.
-
Download the datasets used in the experiments:
-
Modify the
.iniconfiguration files from this folder to match the paths where the datasets are stored on your system as well as the paths where the methods are installed. Run the evaluation as:python -m evalne ./experiments/expLP1.ini
NOTE: In order to obtain the results for, e.g. different values of the embedding dimensionality, the
conf file expLP1.ini has to be modified accordingly and the previous command rerun.
NOTE: For AROPE, VERSE and the GEM library, special main.py files are required in order to run the
evaluation through EvalNE. Once these methods are installed, the corresponding main file has to be added
to the root folder of the method and called from the .ini configuration file. These main.py files are
located in a main_files folder.
If you have found our research useful, please consider citing our paper, which is also available on arxiv:
@INPROCEEDINGS{9260030,
author={A. C. {Mara} and J. {Lijffijt} and T. d. {Bie}},
booktitle={2020 IEEE 7th International Conference on Data Science and Advanced Analytics (DSAA)},
title={Benchmarking Network Embedding Models for Link Prediction: Are We Making Progress?},
year={2020},
pages={138-147},
doi={10.1109/DSAA49011.2020.00026}}