KISTI 기계학습 데이터 공유·활용 서비스

• The corpus for extracting metadata from multilingual journal references.
• The corpus contains the tokens of reference string with IOB labels corresponding to metadata field types.
• The file named as “automatic_inspection_data(train_set).txt” was constructed through the fully automatic inspection process. It contains a total of 3,680,620 labeled references, and was used as training set for training our BERT-based parsing model in our study.
• The file named as “manual_insepction_data_(validation_set).txt” was developed through the manual inspection process by several annotators. It contains a total of 63,878 labeled references, and was used as validation set.
• The file named as “manual_inspection_data_(test_set).txt” was built through the manual inspection process. It contains a total of 71,489 labeled references, and was used as test set.
• In the files, each reference is separated by a newline character. In each first line of references, a unique identifier and language type are listed.

HGJBBV_2020_v21n2_16_005	 English
Jou	B-AUT
,	I-AUT
<sp>	I-AUT
J	I-AUT
.	I-AUT
<sp>	I-AUT
H	I-AUT
.	I-AUT
<sp>	I-AUT
et	I-AUT
<sp>	I-AUT
al	I-AUT
.	O
<sp>	O
Plant	B-TIT
<sp>	I-TIT
growth	I-TIT
<sp>	I-TIT
absorption	I-TIT
<sp>	I-TIT
spectrum	I-TIT
<sp>	I-TIT
mimicking	I-TIT
<sp>	I-TIT
light	I-TIT
<sp>	I-TIT
sources	I-TIT
.	O
<sp>	O
Materials	B-JOU
<sp>	O
8	B-VOL
,	O
<sp>	O
5265	B-PAGE
-	I-PAGE
5275	I-PAGE
<sp>	O
(	O
2015	B-YEAR
)	O
.	O