edfvalim / raea-model-for-entity-alignment

The Pytorch implementation of RAEA model, corresponding to the World Wide Wed 2023 paper "Cross-platform Product Matching based on Entity Alignment of Knowledge Graph with RAEA model", which achieves competive performance on entity alignment in DBP15k and DBP100k datasets.

Introduction

This is the Pytorch implementation for our paper at World Wide Web 2023: Cross-platform productmatching based on entity alignment of knowledge graph with raea model. https://doi.org/10.1007/s11280-022-01134-y

Run RAEA

Let's take experiments on the DBP15k/zh_en as an example.

First pre-train the encoder on the DBP15k/zh_en: python sbert_simcse.py --dataset DBP15k/zh_en

Next, train the RAEA model on the DBP15k/zh_en dataset: python train_subgraph.py. The trained models of defferent channels will be saved.

Then, ensemble the channels and valid the performance on the test set: python ensemble_subgraphs.py

The hyperparameters we use are set to the default values.

You can change the value of hyperparameters. For example, the command for training python train_subgraph.py --dataset DBP15k/zh_en --gpu_id 0 --nega_sample_num 15 defines the dataset, gup_id and nega_sample_num

To understand the hyperparameters setting, please refer to "Training Details" part of our paper.

Citation

Please kindly cite our work as follows if you find our paper or codes helpful.

Liu W, Pan J, Zhang X, et al. Cross-platform product matching based on entity alignment of knowledge graph with raea model. World Wide Web (2023). https://doi.org/10.1007/s11280-022-01134-y