/HashKG

Learning to Hash for Efficient Search over Incomplete Knowledge Graphs, ICDM 2019

Primary LanguagePythonMIT LicenseMIT

Learning to Hash for Efficient Search over Incomplete Knowledge Graphs

  • Introduction

    • The code is a first implementation of our paper at ICDM 2019, including the train process of BIO data;
    • Our code is based on GraphQEmbed, you can get the original code and BIO data on that page.

  • Requirements

    • See ./netquery/requirements.txt

  • Running the code

    • Download the data and unzip it at ./ and reach ./netquery/bio/;

    • We transfer vector x to tanh(δx) and gradually increase δ(from 1 to b) in order to bring x closer to the hamming space. In the code, we used parameter "beta" to represent δ and b is set to 20 in our experiment;

    • Run python train.py --lr 0.001 --beta β --pretrain True to use the edge data to train the embedding, where β gradually increases from 1 to b-1;

    • Then run python train.py --lr 0.001 --beta b --pretrain False to use the query data to fine-tune the embedding;

    • Run python test.py --beta b to compare the difference between the original and the hashed model.

  • Other information

    • The code is currently maintained by Yinlin Jiang. You can contact me at yljiang@seu.edu.cn;
    • This code is a part of our full experiment, we are working on to release a more complete version.