New: Add PTransE (EMNLP 2015) code!

We list the result of various methods implemented by ourselves in dateset FB15k and WN18.

FB15k

WN18

We provide FB15k and WN18 datasets used for the task link prediction in data.zip, using the input format required by our codes. The original data can be downloaded from:

FB15k, WN18 are published by "Translating Embeddings for Modeling Multi-relational Data (2013)." [Download]

FB13, WN11 are published by "Reasoning With Neural Tensor Networks for Knowledge Base Completion". [Download]

New York Times Corpus: The data used in relation extraction from text is publish by "Modeling relations and their mentions without labeled text". The data should be obtained from LDC (https://catalog.ldc.upenn.edu/LDC2008T19) first.

FB40k [Download]

Datasets are required in the folder data/ in the following format, containing six files:

• train.txt: training file, format (e1, e2, rel).

• valid.txt: validation file, same format as train.txt

• test.txt: test file, same format as train.txt.

• entity2id.txt: all entities and corresponding ids, one per line.

• relation2id.txt: all relations and corresponding ids, one per line.

• e1_e2.txt: the top-500 entity pairs which are calculated by TransE. [Download]

The codes are in the folder TransE/, TransR/, CTransR/.

Just type make in the folder ./

For training, you need to follow the steps below:

TransE: call the program Train_TransE in folder TransE/

TransH: call the program Train_TransH in folder TransH/

TransR:

• Train the unif method of TransE as initialization.

• Call the program Train_TransR in folder TransR/

CTransR:

• Train the unif method of TransR as initialization.

• Run the bash run.sh with relation number in folder cluster/ to cluster triples in the trainning data. e.g., bash run.sh 10

• Call the program Train_cTransR in folder CTransR/

You can also change the parameters when running Train_TransE, Train_TransR, Train_CTransR.

-size : the embedding size k, d

-rate : learing rate

-method: 0 - unif, 1 - bern

For testing, you need to follow the steps below:

TransR: Call the program Test_TransR with method as parameter in folder TransR/

CTransR: Call the program Test_CTransR with method as parameter in folder CTransR/

It will evaluate on test.txt and report mean rank and Hits@10.

If you use the code, please kindly cite the following paper:

Yankai Lin, Zhiyuan Liu, Maosong Sun, Yang Liu, Xuan Zhu. Learning Entity and Relation Embeddings for Knowledge Graph Completion. The 29th AAAI Conference on Artificial Intelligence (AAAI'15).[pdf]