ADMH-ER

ADMH-ER

Descriptions

Firstly, we try to generate all modal data based on pre-trained models.

1. Text : we obtain the form of representations, i.e., {"name" : representation}
2. Images : we obtain the form of representations, i.e., {"name" : representation}
3. Videos : we obtain the form of representations, i.e., {"name" : representation}

Secondly, we generate the ids of entities and relations. entities : {entities' name : id} relations : {relations' name : id}

Thirdly, we generate the classification labels and linking labels. Note that, we create the groundtruth with the names. For each name, we have four files, i.e., "xxx.class", "xxx.json" and "xxx.link"

---

Datasets

The raw datasets can be found below.

1. DBP15k Datasets

It is available at the following GitHub repository:

• EVA on GitHub

2. MMKB Datasets

It can be accessed via:

• MMKG Dataset on GitHub

Our processing Datasets for MMER can be found below.

1. MMER Datasets

The rough datasets can be available, and more details of them are coming. You can download them from:

• [MMER Datasets on BaiduPan](Link: https://pan.baidu.com/s/1kbUPWreOKUcrwaBXNDjJFA?pwd=heb9 password: heb9)

Citation

@article{Zhou2025ADMHERAD,
  title={ADMH-ER: Adaptive Denoising Multi-Modal Hybrid for Entity Resolution},
  author={Qian Zhou and Wei Chen and Li Zhang and An Liu and Jun Fang and Lei Zhao},
  journal={IEEE Transactions on Knowledge and Data Engineering},
  year={2025},
  volume={37},
  pages={1049-1063},
  url={https://api.semanticscholar.org/CorpusID:275448400}
}