ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper DR based Sentence & SPO Tuple Pair Generation for Open Information Extraction
Cited 0 time in scopus Share share facebook twitter linkedin kakaostory
Authors
Joonyoung Jung, Dong-oh Kang
Issue Date
2023-10
Citation
International Conference on Information and Communication Technology Convergence (ICTC) 2023, pp.1434-1436
Publisher
IEEE
Language
English
Type
Conference Paper
DOI
https://dx.doi.org/10.1109/ICTC58733.2023.10392362
Abstract
This paper presents a method for generating sentence and SPO (Subject-Predicate-Object) tuple pairs based on dependency relations (DR) to train a DNN model for open information extraction. The proposed system extracts sentences from the OLLIE dataset that contain seed tuples in NL-based sentences with SPO tuples, performs dependency parsing on the NL-based sentences, and generates DR-based sentence and SPO tuple pairs. Consequently, approximately 230,000 DR-based sentence and SPO tuple pairs were generated from the 3 million data. To demonstrate the effectiveness of the generated DR-based learning data in natural language processing, experiments were conducted using the BERT model to recognize SPO tuples. The results showed that the performance of SPO tuple extraction was better when using DR-based learning data compared to NL-based learning data. Specifically, the average accuracy for top1, top-3, and top-5 was 0.08, 0.25, and 0.22 higher, respectively, when using DR-based learning data compared to NL-based learning data.
KSP Keywords
Dependency Parsing, Dependency Relations, Learning data, Natural Language Processing, Open information extraction, Tuple extraction