Home

Awesome

<!--<div align="center"> <img src="https://github.com/zjukg/UMAEA/blob/main/IMG/UMAEA2.jpg" alt="Logo" width="600"> </div>-->

🏝️ Rethinking Uncertainly Missing and Ambiguous Visual Modality in Multi-Modal Entity Alignment

license Pytorch arxiv badge ISWC2023

In the face of modality incompleteness, some models succumb to overfitting the modality noise, and exhibit performance oscillations or declines at high modality missing rates. This indicates that the inclusion of additional multi-modal data can sometimes adversely affect EA. To address these challenges, we introduces UMAEA, a robust multi-modal entity alignment approach designed to tackle uncertainly missing and ambiguous visual modalities.

<div align="center"> <img src="https://github.com/zjukg/UMAEA/blob/main/IMG/case.jpg" width="70%" height="auto" /> </div>

πŸ”” News

πŸ”¬ Dependencies

pip install -r requirement.txt

Details

πŸš€ Train

>> cd UMAEA
>> bash run.sh
# Command Details:
# Bash file / GPU / Dataset / Data Split / R_{sa} / R_{img}
# Begin:
# ---------- R_{img} = 0.4 & iter. & w/o CMMI ----------
>> bash run_umaea_00.sh 0 OEA_D_W_15K_V1 norm 0.2 0.4
>> bash run_umaea_00.sh 0 OEA_D_W_15K_V2 norm 0.2 0.4
>> bash run_umaea_00.sh 0 OEA_EN_FR_15K_V1 norm 0.2 0.4
>> bash run_umaea_00.sh 0 OEA_EN_DE_15K_V1 norm 0.2 0.4
>> bash run_umaea_00.sh 0 DBP15K fr_en 0.3 0.4
>> bash run_umaea_00.sh 0 DBP15K ja_en 0.3 0.4
>> bash run_umaea_00.sh 0 DBP15K zh_en 0.3 0.4
# ---------- R_{img} = 0.6 & non-iter. & w/o CMMI ----------
>> bash run_umaea_0.sh 0 OEA_D_W_15K_V1 norm 0.2 0.6
>> bash run_umaea_0.sh 0 OEA_D_W_15K_V2 norm 0.2 0.6
>> bash run_umaea_0.sh 0 OEA_EN_FR_15K_V1 norm 0.2 0.6
>> bash run_umaea_0.sh 0 OEA_EN_DE_15K_V1 norm 0.2 0.6
>> bash run_umaea_0.sh 0 DBP15K fr_en 0.3 0.6
>> bash run_umaea_0.sh 0 DBP15K ja_en 0.3 0.6
>> bash run_umaea_0.sh 0 DBP15K zh_en 0.3 0.6
# --------- R_{img} = 0.1 & non-iter. & w/ CMMI ---------
>> bash run_umaea_012.sh 0 OEA_D_W_15K_V1 norm 0.2 0.1
>> bash run_umaea_012.sh 0 OEA_D_W_15K_V2 norm 0.2 0.1
>> bash run_umaea_012.sh 0 OEA_EN_FR_15K_V1 norm 0.2 0.1
>> bash run_umaea_012.sh 0 OEA_EN_DE_15K_V1 norm 0.2 0.1
>> bash run_umaea_012.sh 0 DBP15K fr_en 0.3 0.1
>> bash run_umaea_012.sh 0 DBP15K ja_en 0.3 0.1
>> bash run_umaea_012.sh 0 DBP15K zh_en 0.3 0.1
# --------- R_{img} = 0.2 & iter. & w/ CMMI ---------
>> bash run_umaea_012012.sh 0 OEA_D_W_15K_V1 norm 0.2 0.2
>> bash run_umaea_012012.sh 0 OEA_D_W_15K_V2 norm 0.2 0.2
>> bash run_umaea_012012.sh 0 OEA_EN_FR_15K_V1 norm 0.2 0.2
>> bash run_umaea_012012.sh 0 OEA_EN_DE_15K_V1 norm 0.2 0.2
>> bash run_umaea_012012.sh 0 DBP15K fr_en 0.3 0.2
>> bash run_umaea_012012.sh 0 DBP15K ja_en 0.3 0.2
>> bash run_umaea_012012.sh 0 DBP15K zh_en 0.3 0.2

πŸ“Œ Tips: you can open the run_umaea_X.sh file for parameter or training target modification.

🎯 Standard Results

$\bf{H@1}$ Performance with the Settings: w/o surface & Non-iterative. We modified part of the MSNEA to involve not using the content of attribute values but only the attribute types themselves (See issues for details):

Method$\bf{DBP15K_{ZH-EN}}$$\bf{DBP15K_{JA-EN}}$$\bf{DBP15K_{FR-EN}}$
MSNEA.609.541.557
EVA.683.669.686
MCLEA.726.719.719
MEAformer.772.764.771
UMAEA.800.801.818

πŸ“š Dataset (MMEA-UMVM)

To create our MMEA-UMVM (uncertainly missing visual modality) datasets, we perform random image dropping on MMEA datasets. Specifically, we randomly discard entity images to achieve varying degrees of visual modality missing, ranging from 0.05 to the maximum $R_{img}$ of the raw datasets with a step of 0.05 or 0.1. Finally, we get a total number of 97 data split as follow:

<div align="center">
Dataset$R_{img}$
$DBP15K_{ZH-EN}$$0.05, 0.1, 0.15, 0.2, 0.3, 0.4, 0.45, 0.5, 0.55, 0.6, 0.7, 0.75, 0.7829~(STD)$
$DBP15K_{JA-EN}$$0.05, 0.1, 0.15, 0.2, 0.3, 0.4, 0.45, 0.5, 0.55, 0.6, 0.7, 0.7032~(STD)$
$DBP15K_{FR-EN}$$0.05, 0.1, 0.15, 0.2, 0.3, 0.4, 0.45, 0.5, 0.55, 0.6, 0.6758~(STD)$
$OpenEA_{EN-FR}$$0.05, 0.1, 0.15, 0.2, 0.3, 0.4, 0.45, 0.5, 0.55, 0.6, 0.7, 0.8, 0.9, 0.95, 1.0~(STD)$
$OpenEA_{EN-DE}$$0.05, 0.1, 0.15, 0.2, 0.3, 0.4, 0.45, 0.5, 0.55, 0.6, 0.7, 0.8, 0.9, 0.95, 1.0~(STD)$
$OpenEA_{D-W-V1}$$0.05, 0.1, 0.15, 0.2, 0.3, 0.4, 0.45, 0.5, 0.55, 0.6, 0.7, 0.8, 0.9, 0.95, 1.0~(STD)$
$OpenEA_{D-W-V2}$$0.05, 0.1, 0.15, 0.2, 0.3, 0.4, 0.45, 0.5, 0.55, 0.6, 0.7, 0.8, 0.9, 0.95, 1.0~(STD)$
</div>

πŸ“ Download:

import pickle
zh_images = pickle.load(open("eva_image_resources/dbp15k/zh_dbp15k_link_img_dict_full.pkl",'rb'))
print(en_images["http://zh.dbpedia.org/resource/ι¦™ζΈ―ζœ‰η·šι›»θ¦–"].size)
ROOT
β”œβ”€β”€ data
β”‚Β Β  └── mmkg
└── code
 Β Β  └── UMAEA

Code Path

<details> <summary>πŸ‘ˆ πŸ”Ž Click</summary>
UMAEA
β”œβ”€β”€ config.py
β”œβ”€β”€ main.py
β”œβ”€β”€ requirement.txt
β”œβ”€β”€ run.sh
β”œβ”€β”€ run_umaea_00.sh
β”œβ”€β”€ run_umaea_012012.sh
β”œβ”€β”€ run_umaea_012.sh
β”œβ”€β”€ run_umaea_0.sh
β”œβ”€β”€ model
β”‚Β Β  β”œβ”€β”€ __init__.py
β”‚Β Β  β”œβ”€β”€ layers.py
β”‚Β Β  β”œβ”€β”€ Tool_model.py
β”‚Β Β  β”œβ”€β”€ UMAEA_loss.py
β”‚Β Β  β”œβ”€β”€ UMAEA.py
β”‚Β Β  └── UMAEA_tools.py
β”œβ”€β”€ src
β”‚Β Β  β”œβ”€β”€ data.py
β”‚Β Β  β”œβ”€β”€ __init__.py
β”‚Β Β  └── utils.py
β”œβ”€β”€ torchlight
β”‚Β Β  β”œβ”€β”€ __init__.py
β”‚Β Β  β”œβ”€β”€ logger.py
β”‚Β Β  β”œβ”€β”€ metric.py
β”‚Β Β  └── utils.py
└── tree.txt
</details>

Data Path

<details> <summary>πŸ‘ˆ πŸ”Ž Click</summary>
mmkg
β”œβ”€β”€ dump
β”œβ”€β”€ DBP15K
β”‚Β Β  β”œβ”€β”€ fr_en
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ent_ids_1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ent_ids_2
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ill_ent_ids
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ training_attrs_1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ training_attrs_2
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ triples_1
β”‚Β Β  β”‚Β Β  └── triples_2
β”‚Β Β  β”œβ”€β”€ ja_en
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ent_ids_1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ent_ids_2
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ill_ent_ids
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ training_attrs_1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ training_attrs_2
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ triples_1
β”‚Β Β  β”‚Β Β  └── triples_2
β”‚Β Β  β”œβ”€β”€ translated_ent_name
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ dbp_fr_en.json
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ dbp_ja_en.json
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ dbp_zh_en.json
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ srprs_de_en.json
β”‚Β Β  β”‚Β Β  └── srprs_fr_en.json
β”‚Β Β  └── zh_en
β”‚Β Β      β”œβ”€β”€ ent_ids_1
β”‚Β Β      β”œβ”€β”€ ent_ids_2
β”‚Β Β      β”œβ”€β”€ ill_ent_ids
β”‚Β Β      β”œβ”€β”€ training_attrs_1
β”‚Β Β      β”œβ”€β”€ training_attrs_2
β”‚Β Β      β”œβ”€β”€ triples_1
β”‚Β Β      └── triples_2
β”œβ”€β”€ OpenEA
β”‚Β Β  β”œβ”€β”€ OEA_D_W_15K_V1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ent_ids_1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ent_ids_2
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ill_ent_ids
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ rel_ids
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ training_attrs_1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ training_attrs_2
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ triples_1
β”‚Β Β  β”‚Β Β  └── triples_2
β”‚Β Β  β”œβ”€β”€ OEA_D_W_15K_V2
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ent_ids_1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ent_ids_2
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ill_ent_ids
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ rel_ids
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ training_attrs_1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ training_attrs_2
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ triples_1
β”‚Β Β  β”‚Β Β  └── triples_2
β”‚Β Β  β”œβ”€β”€ OEA_EN_DE_15K_V1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ent_ids_1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ent_ids_2
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ill_ent_ids
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ rel_ids
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ training_attrs_1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ training_attrs_2
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ triples_1
β”‚Β Β  β”‚Β Β  └── triples_2
β”‚Β Β  β”œβ”€β”€ OEA_EN_FR_15K_V1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ent_ids_1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ent_ids_2
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ ill_ent_ids
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ rel_ids
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ training_attrs_1
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ training_attrs_2
β”‚Β Β  β”‚Β Β  β”œβ”€β”€ triples_1
β”‚Β Β  β”‚Β Β  └── triples_2
β”‚Β Β  └── pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.05.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.15.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.1.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.2.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.3.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.45.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.4.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.55.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.5.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.6.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.75.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.7.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.8.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.95.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict_0.9.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V1_id_img_feature_dict.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.05.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.15.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.1.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.2.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.3.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.45.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.4.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.55.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.5.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.6.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.75.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.7.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.8.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.95.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict_0.9.pkl
β”‚Β Β      β”œβ”€β”€ OEA_D_W_15K_V2_id_img_feature_dict.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.05.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.15.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.1.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.2.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.3.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.45.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.4.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.55.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.5.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.6.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.75.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.7.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.8.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.95.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict_0.9.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_DE_15K_V1_id_img_feature_dict.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.05.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.15.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.1.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.2.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.3.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.45.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.4.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.55.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.5.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.6.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.75.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.7.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.8.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.95.pkl
β”‚Β Β      β”œβ”€β”€ OEA_EN_FR_15K_V1_id_img_feature_dict_0.9.pkl
β”‚Β Β      └── OEA_EN_FR_15K_V1_id_img_feature_dict.pkl
β”œβ”€β”€ pkls
β”‚Β Β  β”œβ”€β”€ fr_en_GA_id_img_feature_dict_0.05.pkl
β”‚Β Β  β”œβ”€β”€ fr_en_GA_id_img_feature_dict_0.15.pkl
β”‚Β Β  β”œβ”€β”€ fr_en_GA_id_img_feature_dict_0.1.pkl
β”‚Β Β  β”œβ”€β”€ fr_en_GA_id_img_feature_dict_0.2.pkl
β”‚Β Β  β”œβ”€β”€ fr_en_GA_id_img_feature_dict_0.3.pkl
β”‚Β Β  β”œβ”€β”€ fr_en_GA_id_img_feature_dict_0.45.pkl
β”‚Β Β  β”œβ”€β”€ fr_en_GA_id_img_feature_dict_0.4.pkl
β”‚Β Β  β”œβ”€β”€ fr_en_GA_id_img_feature_dict_0.55.pkl
β”‚Β Β  β”œβ”€β”€ fr_en_GA_id_img_feature_dict_0.5.pkl
β”‚Β Β  β”œβ”€β”€ fr_en_GA_id_img_feature_dict_0.6.pkl
β”‚Β Β  β”œβ”€β”€ fr_en_GA_id_img_feature_dict_0.7.pkl
β”‚Β Β  β”œβ”€β”€ fr_en_GA_id_img_feature_dict.pkl
β”‚Β Β  β”œβ”€β”€ ja_en_GA_id_img_feature_dict_0.05.pkl
β”‚Β Β  β”œβ”€β”€ ja_en_GA_id_img_feature_dict_0.15.pkl
β”‚Β Β  β”œβ”€β”€ ja_en_GA_id_img_feature_dict_0.1.pkl
β”‚Β Β  β”œβ”€β”€ ja_en_GA_id_img_feature_dict_0.2.pkl
β”‚Β Β  β”œβ”€β”€ ja_en_GA_id_img_feature_dict_0.3.pkl
β”‚Β Β  β”œβ”€β”€ ja_en_GA_id_img_feature_dict_0.45.pkl
β”‚Β Β  β”œβ”€β”€ ja_en_GA_id_img_feature_dict_0.4.pkl
β”‚Β Β  β”œβ”€β”€ ja_en_GA_id_img_feature_dict_0.55.pkl
β”‚Β Β  β”œβ”€β”€ ja_en_GA_id_img_feature_dict_0.5.pkl
β”‚Β Β  β”œβ”€β”€ ja_en_GA_id_img_feature_dict_0.6.pkl
β”‚Β Β  β”œβ”€β”€ ja_en_GA_id_img_feature_dict.pkl
β”‚Β Β  β”œβ”€β”€ zh_en_GA_id_img_feature_dict_0.05.pkl
β”‚Β Β  β”œβ”€β”€ zh_en_GA_id_img_feature_dict_0.15.pkl
β”‚Β Β  β”œβ”€β”€ zh_en_GA_id_img_feature_dict_0.1.pkl
β”‚Β Β  β”œβ”€β”€ zh_en_GA_id_img_feature_dict_0.2.pkl
β”‚Β Β  β”œβ”€β”€ zh_en_GA_id_img_feature_dict_0.3.pkl
β”‚Β Β  β”œβ”€β”€ zh_en_GA_id_img_feature_dict_0.45.pkl
β”‚Β Β  β”œβ”€β”€ zh_en_GA_id_img_feature_dict_0.4.pkl
β”‚Β Β  β”œβ”€β”€ zh_en_GA_id_img_feature_dict_0.55.pkl
β”‚Β Β  β”œβ”€β”€ zh_en_GA_id_img_feature_dict_0.5.pkl
β”‚Β Β  β”œβ”€β”€ zh_en_GA_id_img_feature_dict_0.6.pkl
β”‚Β Β  β”œβ”€β”€ zh_en_GA_id_img_feature_dict_0.75.pkl
β”‚Β Β  β”œβ”€β”€ zh_en_GA_id_img_feature_dict_0.7.pkl
β”‚Β Β  └── zh_en_GA_id_img_feature_dict.pkl
└── UMAEA
    └── save
</details>

🀝 Cite:

Please condiser citing this paper if you use the code or data from our work. Thanks a lot :)

@inproceedings{DBLP:conf/semweb/ChenGFZCPLCZ23,
  author       = {Zhuo Chen and
                  Lingbing Guo and
                  Yin Fang and
                  Yichi Zhang and
                  Jiaoyan Chen and
                  Jeff Z. Pan and
                  Yangning Li and
                  Huajun Chen and
                  Wen Zhang},
  title        = {Rethinking Uncertainly Missing and Ambiguous Visual Modality in Multi-Modal
                  Entity Alignment},
  booktitle    = {{ISWC}},
  series       = {Lecture Notes in Computer Science},
  volume       = {14265},
  pages        = {121--139},
  publisher    = {Springer},
  year         = {2023}
}

πŸ’‘ Acknowledgement