๋ณธ๋ฌธ ๋ฐ”๋กœ๊ฐ€๊ธฐ

Covid-19:ํŠธ์œ„ํ„ฐ ๊ฐ์ •๋ถ„์„

ํŠธ์œ„ํ„ฐ ๋ฐ์ดํ„ฐ KoBERT ๊ฐ์ •๋ถ„๋ฅ˜ ๊ฒฐ๊ณผ์ •๋ฆฌ
Projects/COVID-19 analysis ํŠธ์œ„ํ„ฐ ๋ฐ์ดํ„ฐ KoBERT ๊ฐ์ •๋ถ„๋ฅ˜ ๊ฒฐ๊ณผ์ •๋ฆฌ 1.KoBERT ์†Œ๊ฐœ KoBERT๋Š” BERT ์˜ ํ•œ๊ตญ์–ด๋ฒ„์ „์ž…๋‹ˆ๋‹ค. BERT(Bidirectional Encoder Representations from Transformers)๋Š” ๊ตฌ๊ธ€์ด ๊ณต๊ฐœํ•œ ์ธ๊ณต์ง€๋Šฅ(AI) ์–ธ์–ด๋ชจ๋ธ์ธ๋ฐ์š”, ์ผ๋ถ€ ์„ฑ๋Šฅ ํ‰๊ฐ€์—์„œ ์ธ๊ฐ„๋ณด๋‹ค ๋” ๋†’์€ ์ •ํ™•๋„๋ฅผ ๋ณด์ด๋ฉฐ 2018๋…„ ๋ง์— ์ž์—ฐ ์–ธ์–ด ์ฒ˜๋ฆฌ(NLP)์—์„œ SOTA๋ฅผ ๋‹ฌ์„ฑํ•œ ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค. BERT์˜ ํŠน์ง•์œผ๋กœ๋Š” ์„ธ ๊ฐ€์ง€๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. - ์–ธ์–ดํ‘œํ˜„ ์‚ฌ์ „ํ•™์Šต์˜ ์ƒˆ๋กœ์šด ๋ฐฉ๋ฒ• Wikipedia ๋‚˜ BooksCorpus์™€ ๊ฐ™์€ ๋Œ€์šฉ๋Ÿ‰์˜ ๋ผ๋ฒจ๋ง ๋˜์–ด ์žˆ์ง€ ์•Š์€ ๋ฐ์ดํ„ฐ(์ •๋‹ต์ด ์—†๋Š”)๋กœ ๋ชจ๋ธ์„ pretraining ์‹œํ‚จ ํ›„, ํŠน์ • task๋ฅผ ๊ฐ€์ง€๊ณ  ์žˆ๋Š” labeled data๋กœ transfer learning์„ ํ•˜๋Š” ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค. - Bidirectional ..
[๋ฐ์ดํ„ฐ์…‹] Sentiment-analysis dataset
Projects/COVID-19 analysis [๋ฐ์ดํ„ฐ์…‹] Sentiment-analysis dataset http://nlpprogress.com/english/sentiment_analysis.html https://paperswithcode.com/task/sentiment-analysis sentiment analysis english dataset ์„ ์ •๋ฆฌํ•ด๋ณด์•˜์Šต๋‹ˆ๋‹ค. SST์™€ Yelp dataset๋Š” sentimentrk 5๋‹จ๊ณ„๋กœ ๋‚˜๋‰œ sentimtiment dataset์ด ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. Sem-Eval-2017 Task4 ๋ฐ์ดํ„ฐ์…‹์€ Subtask A ์— ๊ธ์ •, ๋ถ€์ •, ์ค‘๋ฆฝ ๋ฐ์ดํ„ฐ์…‹์ด ์žˆ์Šต๋‹ˆ๋‹ค. ๋‚˜๋จธ์ง€ ๋ฐ์ดํ„ฐ์…‹๋“ค์€ ๊ธ์ •๊ณผ ๋ถ€์ •์œผ๋กœ๋งŒ ๋ถ„๋ฅ˜๋˜์–ด์žˆ์Šต๋‹ˆ๋‹ค. ๊ฐ ๋ฐ์ดํ„ฐ์…‹๋“ค ๋งˆ๋‹ค SOTA ๋ชจ๋ธ๊ณผ ์š”์ฆ˜ ๊ฐ€์žฅ ํ•ซํ•œ BERT ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ์ •๋ฆฌํ–ˆ์Šต๋‹ˆ๋‹ค.

ํ‹ฐ์Šคํ† ๋ฆฌํˆด๋ฐ”