Building a Massive Corpus for Named Entity Recognition using Free Open Data Sources

Menezes, Daniel Specht; Savarese, Pedro; Milidiú, Ruy Luiz

doi:10.48550/arxiv.1908.05758

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition, we note that satisfactory results of NER rely heavily on a large quantity of hand-annotated data, which is often costly in terms of time and resources spent. Therefore, the adoption of semi-supervised and/or unsupervised learning methods could reduce the need for manual annotation 51,52 by incorporating unlabeled data to improve the accuracy and deal with more battery properties and evaluate it against standardized IE benchmarks.…”

Section: Resultsmentioning

confidence: 99%

LIBAC: An Annotated Corpus for Automated “Reading” of the Lithium-Ion Battery Research Literature

et al. 2023

View full text Add to dashboard Cite

The lithium-ion battery (LIB) research literature has increased very rapidly of late. While this is an immense source of valuable knowledge and facts for the community, these are also partly "buried" in the literature. To truly make the most possible use of the information available and automate "reading", special tools are required. Named entity recognition (NER) is one such tool, which uses supervised machine learning for information extraction. To enable efficient NER, however, a large and highquality annotated corpus is crucial. Here, we report on our generated, semi-automatically annotated lithium-ion battery annotated corpus, "LIBAC", for 28 different entities of LIBs, which was used for training and evaluating Tok2vec and Transformer-based models, resulting in high general accuracies for these with F 1 -scores of 81 and 83%, respectively. LIBAC itself was created from 6985 paragraphs randomly chosen from ca. 11,000 LIB research papers and contains 73,300 annotated spans (627,428 tokens). This is the prime stepping-stone needed to develop a large-scale information extraction system designed for the LIB research literature.

show abstract

Section: Resultsmentioning

confidence: 99%