DAEA: Enhancing Entity Alignment in Real-World Knowledge Graphs Through Multi-Source Domain Adaptation

This is code and datasets for DAEA

Dependencies

• Python 3 (tested on 3.9.0)
• Pytorch (tested on 2.2.1)
• transformers (tested on 2.1.1)
• torch_geometric (tested on 2.5.3)

Dataset

DBP15K and Real-World Data

All the data can be downloaded from https://drive.google.com/file/d/1OXzwPve7PT1jRHfdZajNro2MYM4I7Lc0/view?usp=drive_link.

Initial DBP15K datasets are from JAPE(https://github.com/nju-websoft/JAPE). Initial Real-World datasets are from https://github.com/EnsiyehRaoufi/Create_Input_Data_to_EA_Models.

Description data

• data/dbp15k/2016-10-des_dict: A dictionary storing entity descriptions, which can be loaded by pickle.load(). The description of the entity is extracted from DBpedia(https://wiki.dbpedia.org/downloads-2016-10)
• data/agrold_en/2024-06-03-agrold: A dictionary storing entity descriptions, which can be loaded by pickle.load(). The description of the entity is extracted from https://github.com/EnsiyehRaoufi/Create_Input_Data_to_EA_Models.
• data/doremus_en/2024-02-12-doremus: A dictionary storing entity descriptions, which can be loaded by pickle.load(). The description of the entity is extracted from https://github.com/EnsiyehRaoufi/Create_Input_Data_to_EA_Models.

How to Run

The model runs in three steps:

1. Multi-Source KGs selection

The source and target datasets path can be changed in multisource/param.py.

cd multisource/
python main.py

2. Fine-tune Basic BERT Unit followed https://github.com/kosugi11037/bert-int

To fine-tune the Basic BERT Unit, use:

cd basic_bert_unit/
python main.py

Note that basic_bert_unit/Param.py is the config file.

The obtained Basic BERT Unit and some other data will be stored in: ../Save_model

3. Run BERT-based Interaction Model followed https://github.com/kosugi11037/bert-int

(Note that when running the BERT-based Interaction model, the parameters of the Basic BERT Unit model will be fixed.)

To extract the similarity features and run the BERT-base Interaction Model, use:

cd ../interaction_model/
python clean_attribute_data.py
python get_entity_embedding.py
python get_attributeValue_embedding.py
python get_neighView_and_desView_interaction_feature.py
python get_attributeView_interaction_feature.py
python interaction_model.py

Or directly use:

cd ../interaction_model/
bash run.sh

Note that interaction_model/Param.py is the config file.

Acknowledgement

We extend our appreciation to: BERT-INT (https://github.com/kosugi11037/bert-int), real-world datasets https://github.com/EnsiyehRaoufi/Create_Input_Data_to_EA_Models, transfer learning library https://github.com/jindongwang/transferlearning/tree/master.