OpenBG500

Information

OpenBG500 is an open chinese E-commerce and bussiness knowledge graph dataset contained 500 relations. This dataset is refined from the OpenBG, a million-scale multi-modal dataset evolving products and consumption demands in a unified schema. AliOpenKG500 is developed for several knowledge graph embedding evaluations.

The dataset splits all data into 3 parts. Base statistical information is shown in the table below.

#Relation	#Entity	#Train (opened)	#Valid (opened)	#Test
500	249,743	1242550	5000	5000

Data

OpenBG500 is available at Google Drive and Baidu Netdisk(password: 78fw). The main derectory of the dataset is as follows.

OpenBG500
├── OpenBG500_train.tsv 			# Training set
├── OpenBG500_dev.tsv 				# Validation set
├── OpenBG500_test.tsv 			    # Test set
├── OpenBG500_entity2text.tsv 		# Description of entities in Chinese
├── OpenBG500_relation2text.tsv 	# Description of relations in Chinese
└── OpenBG500_example_pred.tsv 	    # Submit example

Usage

Format

• Triples

# OpenBG500_train.tsv/OpenBG500_dev.tsv
Head<\t>Relation<\t>Tail<\n>

• Description of entities/relations in Chinese

# OpenBG500_entity2text.tsv/OpenBG500_relation2text.tsv
Entity(Relation)<\t>Description of entitie(relation)<\n>

• Test and submit

# For OpenBG500_test.tsv, participants are required to predict 10 Tails for one instance. OpenBG500_example_pred.tsv is a submit example.
Head<\t>Relation<\n>

# OpenBG500_example_pred.tsv
Head<\t>Relation<\t>Tail 1<\t>Tail 2<\t>...<\t>Tail 10<\n>

Check the data

$ head -n 3 OpenBG500_train.tsv
ent_135492      rel_0352        ent_015651
ent_020765      rel_0448        ent_214183
ent_106905      rel_0418        ent_121073

Read the datasets

1. Read the original data:

with open('OpenBG500_train.tsv', 'r') as fp:
    data = fp.readlines()
    train = [line.strip('\n').split('\t') for line in data]
    _ = [print(line) for line in train[:2]]
    # ['ent_135492', 'rel_0352', 'ent_015651']
    # ['ent_020765', 'rel_0448', 'ent_214183']

2. Get the map of Entity(Relatioin)-Description: ent2text and rel2text:

with open('OpenBG500_entity2text.tsv', 'r') as fp:
    data = fp.readlines()
    lines = [line.strip('\n').split('\t') for line in data]
    _ = [print(line) for line in lines[:2]]
    # ['ent_101705', '短袖T恤']
    # ['ent_116070', '套装']

ent2text = {line[0]: line[1] for line in lines}

with open('OpenBG500_relation2text.tsv', 'r') as fp:
    data = fp.readlines()
    lines = [line.strip().split('\t') for line in data]
    _ = [print(line) for line in lines[:2]]
    # ['rel_0418', '细分市场']
    # ['rel_0290', '关联场景']

rel2text = {line[0]: line[1] for line in lines}

3. Transfer the data to description:

train = [[ent2text[line[0]],rel2text[line[1]],ent2text[line[2]]] for line in train]
_ = [print(line) for line in train[:2]]
# ['苦荞茶', '外部材质', '苦荞麦']
# ['精品三姐妹硬糕', '口味', '原味硬糕850克【10包40块糕】']

Submit in Alibaba TIANCHI

OpenBG Benchmark：Large Scale Open Business Knowledge Graph Benchmark is a benchmark open for a long time. Welcome to submit your result of OpenBG500.

Baseline result

We do some baseline method on this dataset. TransE, DistMult and ComplEx result are based on OpenKE toolkit, KG-BERT and GenKGC results are based our code.

Method	Hits@1	Hits@3	Hits@10
TransE	0.207	0.340	0.531
DistMult	0.049	0.088	0.216
ComplEx	0.053	0.120	0.266
KG-BERT	0.023	0.049	0.241
GenKGC	0.203	0.280	0.351