MNE: Multi-relational Network Embedding

This is the codes and data for MNE model in our paper "A Structural Representation Learning for Multi-relational Networks".

If you would like to acknowledge our efforts, please cite the following paper:

@inproceedings{ijcai2017-565,
author    = {Lin Liu, Xin Li, William K. Cheung, Chengcheng Xu},
title     = {A Structural Representation Learning for Multi-relational Networks},
booktitle = {Proceedings of the Twenty-Sixth International Joint Conference on
              Artificial Intelligence, {IJCAI-17}},
pages     = {4047--4053},
year      = {2017},
doi       = {10.24963/ijcai.2017/565},
url       = {https://doi.org/10.24963/ijcai.2017/565},
}

Prerequisites

C++

python 3

sklearn

Usage

There are five files/folder (some may be zipped to meet the space limit):

• MNE.cpp: The MNE model for embedding
• logistic_classification.py: The classifier for classification
• logistic_link_prediction.py: The classifier for link prediction
• FB15k: The dataset case from FreeBase
• WN18: The dataset case from WordNet

Run for classification

./mne -trainset data/all_train.txt -outputdir output/ -entity2id data/entity2id.txt -relation2id data/relation2id.txt
python logistic_classification.py -entityVec output/entity_vec.txt -relationVec output/relation_vec.txt -trainLink data/train_link.txt -result result/testresult_MNE.txt

Run for link prediction

./mne -trainset data/train.txt -outputdir output/ -entity2id data/entity2id.txt -relation2id data/relation2id.txt
python logistic_link_prediction.py -entityVec output/entity_vec.txt -relationVec output/relation_vec.txt -trainLink data/train_link.txt -testLink data/test_link.txt -result result/testresult_MNE.txt

Note

• The output files (learned embeddings) will be stored in the -outputdir directory during embedding training.
• Please make the -result directory before run the program, and the test results will be stored in the -result directory.
• the all_train.txt in the run case means the file which contans all triples in dataset, equalling train.txt + valid.txt + test.txt.