【演講】2022.12.14The association between annotation consistency and model enhancement: a case study in BioNER task
Start Date
2022-12-14 13:00:00
End Date
2022-12-14 15:00:00
Introduction
The association between annotation consistency and model enhancement: a case study in BioNER taskrnrn日期 : 2022.12.14 星期三rn時間 : 13:10-15:00rn地點 : 台北醫學大學大安校區B2 （B201）rnrnAbstractrnrnNumerous text inputs are required to establish machine system for natural language processing （NLP）. The pre-labeled documents serve as datasets or corpora for system compilation. Therefore, the annotation quality of included corpus would mainly affect the machine efficacy.rnrn In this topic, we briefly introduce the concept of annotation”, which is the foundation of text mining. The well-known Bio-NER corpus, JNLPBA, is used as an actual instance, and the reports from several systems on JNLPBA are provided to illustrate the phenomena of annotation inconsistency. Moreover, we revise JNLPBA corpus, and the corpus quality is measured by inter-annotator agreement （IAA） analysis. Lastly, the comparisons of system training on original/revised JNLPBA are applied to prove the association between annotation consistency and system performance.
