[๋…ผ๋ฌธ๋ฆฌ๋ทฐ] It’s Not Just Size That Matters: Small Language Models Are Also Few-Shot Learners
ยท
Artificial_Intelligence๐Ÿค–/Natural Language Processing
Schick, Timo, and Hinrich Schütze. "Exploiting Cloze-Questions for Few-Shot Text Classification and Natural Language Inference." Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. 2021. Schick, Timo, and Hinrich Schütze. "It’s Not Just Size That Matters: Small Language Models Are Also Few-Shot Learners." Proceedings of the 20..
Text Similarity, Semantic Similarity
ยท
Artificial_Intelligence๐Ÿค–/Natural Language Processing
ํ…์ŠคํŠธ ์œ ์‚ฌ๋„ ์ฝ”์‚ฌ์ธ ์œ ์‚ฌ๋„ (Cosine Similarity) -> ๋‘ ๊ฐœ์˜ ๋ฒกํ„ฐ ๊ฐ’์˜ Cos ๊ฐ๋„ ์œ ํด๋ฆฌ๋””์–ธ ์œ ์‚ฌ๋„ (Euclidean Similarity) -> ๋‘ ๊ฐœ์˜ ์  ์‚ฌ์ด์˜ ๊ฑฐ๋ฆฌ = L2 ๊ฑฐ๋ฆฌ ๋งจํ•˜ํƒ„ ์œ ์‚ฌ๋„ (Menhattan Similarity) -> ์‚ฌ๊ฐ ๊ฒฉ์ž ์ตœ๋‹จ ๊ฑฐ๋ฆฌ = L1 ๊ฑฐ๋ฆฌ ์ž์นด๋“œ ์œ ์‚ฌ๋„ (Jaccard Similarity) -> ๊ต์ง‘ํ•ฉ๊ณผ ํ•ฉ์ง‘ํ•ฉ์˜ ํฌ๊ธฐ๋กœ ๊ณ„์‚ฐ ๋‘ ๋ฌธ์žฅ์ด ์ฃผ์–ด์กŒ์„ ๋•Œ, ๋‘ ๋ฌธ์žฅ์ด ์„œ๋กœ ์–ผ๋งˆ๋‚˜ ์œ ์‚ฌํ•œ์ง€ ๋‚˜ํƒ€๋‚ด์ฃผ๋Š” ๊ธฐ๋ฒ• ์•„๋ž˜์—์„œ ์ž…๋ ฅ๊ฐ’์œผ๋กœ ๋ฐ›๋Š” Sentences๋Š” ["Hello World", "Hello Word"] ํ˜•์‹์ด๋‹ค. ### ์ฝ”์‚ฌ์ธ ์œ ์‚ฌ๋„ ### def cos_performance(sentences) : tfidf_vectorizer = TfidfVecto..
Count-Base Word Representation
ยท
Artificial_Intelligence๐Ÿค–/Natural Language Processing
์นด์šดํŠธ ๊ธฐ๋ฐ˜์˜ ๋‹จ์–ด ํ‘œํ˜„์ด๋ž€ ์–ด๋–ค ๊ธ€์˜ ๋ฌธ๋งฅ ์•ˆ์— ๋‹จ์–ด๊ฐ€ ๋™์‹œ์— ๋“ฑ์žฅํ•˜๋Š” ํšŸ์ˆ˜๋ฅผ ์„ธ๋Š” ๋ฐฉ๋ฒ•์ž…๋‹ˆ๋‹ค. ๋™์‹œ ๋“ฑ์žฅ ํšŸ์ˆ˜๋ฅผ ํ•˜๋‚˜์˜ ํ–‰๋ ฌ๋กœ ๋‚˜ํƒ€๋‚ธ ๋’ค, ๊ทธ ํ–‰๋ ฌ์„ ์ˆ˜์น˜ํ™”ํ•ด์„œ ๋‹จ์–ด ๋ฒกํ„ฐ๋กœ ๋งŒ๋“œ๋Š” ๋ฐฉ๋ฒ•์„ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ์‹์ž…๋‹ˆ๋‹ค. ํ…์ŠคํŠธ๋ฅผ ์œ„์™€ ๊ฐ™์€ ๋ฐฉ์‹์œผ๋กœ ์ˆ˜์น˜ํ™”ํ•˜๋ฉด, ํ†ต๊ณ„์ ์ธ ์ ‘๊ทผ ๋ฐฉ๋ฒ•์„ ํ†ตํ•ด ์—ฌ๋Ÿฌ ๋ฌธ์„œ๋กœ ์ด๋ฃจ์–ด์ง„ ํ…์ŠคํŠธ ๋ฐ์ดํ„ฐ๊ฐ€ ์žˆ์„ ๋•Œ ์–ด๋–ค ๋‹จ์–ด๊ฐ€ ํŠน์ • ๋ฌธ์„œ๋‚ด์—์„œ ์–ผ๋งˆ๋‚˜ ์ค‘์š”ํ•œ ๊ฒƒ์ธ์ง€๋ฅผ ๋‚˜ํƒ€๋‚ด๊ฑฐ๋‚˜, ๋ฌธ์„œ์˜ ํ•ต์‹ฌ์–ด ์ถ”์ถœ, ๊ฒ€์ƒ‰ ์—”์ง„์—์„œ ๊ฒ€์ƒ‰ ๊ฒฐ๊ณผ์˜ ์ˆœ์œ„ ๊ฒฐ์ •, ๋ฌธ์„œ๋“ค ๊ฐ„์˜ ์œ ์‚ฌ๋„ ๋“ฑ์˜ ์šฉ๋„๋กœ ์‚ฌ์šฉ๊ฐ€๋Šฅํ•ฉ๋‹ˆ๋‹ค. ๊ฐ ๋‹จ์–ด์— 1๋ฒˆ, 2๋ฒˆ, 3๋ฒˆ ๋“ฑ๊ณผ ๊ฐ™์€ ์ˆซ์ž๋ฅผ ๋งตํ•‘(mapping)ํ•˜์—ฌ ๋ถ€์—ฌํ•œ๋‹ค๋ฉด ์ด๋Š” ๊ตญ์†Œ ํ‘œํ˜„ ๋ฐฉ๋ฒ•์— ํ•ด๋‹น๋ฉ๋‹ˆ๋‹ค. ๋ฐ˜๋ฉด, ๋ถ„์‚ฐ ํ‘œํ˜„ ๋ฐฉ๋ฒ•์˜ ํ•ด๋‹น ๋‹จ์–ด๋ฅผ ํ‘œํ˜„ํ•˜๊ธฐ ์œ„ํ•ด ์ฃผ๋ณ€ ๋‹จ์–ด๋ฅผ ์ฐธ๊ณ ํ•ฉ๋‹ˆ๋‹ค. puppy(๊ฐ•์•„..
Natural Language Processing with Disaster Tweets
ยท
Artificial_Intelligence๐Ÿค–/Natural Language Processing
Natural Language Processing with Disaster Tweets Predict which Tweets are about real disasters and which ones are not https://www.kaggle.com/c/nlp-getting-started Natural Language Processing with Disaster Tweets | Kaggle www.kaggle.com NLP ๊ณต๋ถ€๋ฅผ ํ•˜๋ฉด์„œ ์ดˆ๊ธฐ ๋…ผ๋ฌธ๋ถ€ํ„ฐ ํ•˜๋‚˜์”ฉ ๋ณด๋ฉด์„œ ์ž‘์„ฑํ•ด๋ณด๊ณ , ์ตœ์‹  ํŠธ๋ Œ๋“œ๋ฅผ ๊ณต๋ถ€ํ•ด๊ฐ€๋ฉด์„œ, ์ง์ ‘ ๋ฐ์ดํ„ฐ ์ฒ˜๋ฆฌ๋ถ€ํ„ฐ ๋ชจ๋ธ์„ ๋Œ๋ ค๋ณด๊ณ , ์ž์—ฐ์–ด๋ฅผ ์–ด๋–ป๊ฒŒ ์ฒ˜๋ฆฌํ•˜๋Š”์ง€ ๊ณผ์ •์„ ์ง์ ‘ ๊ฒฝํ—˜ํ•ด ๋ณด๊ณ  ์‹ถ์—ˆ๋‹ค. ์ฆ‰, NLP ๋ชจ๋ธ์„ ๋Œ๋ฆฌ๊ธฐ ์œ„ํ•œ ์ง์ ‘ ์ฝ”๋”ฉ์„ ํ•˜๊ณ  ์‹ถ์—ˆ๋‹ค. ๊ธฐ์กด์— BERT Model์„ ๊ณต๋ถ€ํ•˜๋ฉด์„œ ..
New NLP Trands
ยท
Artificial_Intelligence๐Ÿค–/Natural Language Processing
Timkey, W. and van Schijndel, M. (2021) → Rogue(์ž‘์€ ๋ช‡๊ฐœ์˜ ์ฐจ์›) ๊ฐœ๋… ์ œ์•ˆ. → rogue๊ฐ€ ๋ชจ๋ธ์„ ์ขŒ์šฐํ•˜๋‹ˆ, ์ด๋ฅผ ์ œ์–ดํ•˜๋Š” postprocessing ํ…Œํฌ๋‹‰ ์ œ์•ˆ Paik, C., Aroca-Ouellette, S., Roncone, A., and Kann, K. (2021) → CoDa(์‚ฌ๋žŒ์ด ์ธ์ง€ ๊ฐ€๋Šฅํ•œ ์ƒ‰์„ ๊ตฌ๋ถ„ํ•˜๊ธฐ ์œ„ํ•œ ๋ฐ์ดํ„ฐ) ๊ตฌ์„ฑ → PLM์˜ ํ•œ๊ณ„ ์ง€์ . (๋ณ‘๋ฐฑํ•˜๊ฒŒ ๋”ฑ ์ด๊ฑฐ๋‹ค! ๋ผ๊ณ  ๋งํ•˜๋Š” ์‚ฌ๋žŒx. ํ…์ŠคํŠธ๋งŒ์œผ๋กœ๋Š” ์ด๋Ÿฌํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ์ธ์ง€ํ•˜๋Š” ๊ฒƒ์— ๋ถ€์กฑํ•จ ๋ฐœ๊ฒฌ. ๋”ฐ๋ผ์„œ ๋‹ค์–‘ํ•œ ํ˜•ํƒœ์˜ ๋ฐ์ดํ„ฐ๋ฅผ ์–ธ์–ด ๋ชจ๋ธ์— ์ ์šฉํ•˜๋Š” ๋ฐฉ๋ฒ• ํƒ๊ตฌ Kalyan, A., Kumar, A., Chandrasekaran, A., Sabharwal, A., and Clark, P. (20..
[๋…ผ๋ฌธ๋ฆฌ๋ทฐ]Efficient Estimation of Word Representations in Vector Space
ยท
Artificial_Intelligence๐Ÿค–/Natural Language Processing
2022.02.20 - [Artificial_Intelligence/Papers] - [๋…ผ๋ฌธ๋ฆฌ๋ทฐ]Distributed Representations of Words and Phrases and their Compositionality [๋…ผ๋ฌธ๋ฆฌ๋ทฐ]Distributed Representations of Words and Phrases and their Compositionality ใ„ดDistributed Representations of Words and Phrases and their Compositionality Mikolov, Tomas, et al. "Distributed representations of words and phrases and their compositionality." Advance..
[๋…ผ๋ฌธ๋ฆฌ๋ทฐ]Distributed Representations of Words and Phrases and their Compositionality
ยท
Artificial_Intelligence๐Ÿค–/Natural Language Processing
Distributed Representations of Words and Phrases and their Compositionality Mikolov, Tomas, et al. "Distributed representations of words and phrases and their compositionality." Advances in neural information processing systems 26 (2013). Abstract (Eng.) The recently introduced continuous Skip-gram model is an efficient method for learning high-quality distributed vector representations that cap..
[๋…ผ๋ฌธ๋ฆฌ๋ทฐ]A Neural Probabilistic Language Model
ยท
Artificial_Intelligence๐Ÿค–/Natural Language Processing
A Neural Probabilistic Language Model Bengio, Yoshua, Réjean Ducharme, and Pascal Vincent. "A neural probabilistic language model." Advances in Neural Information Processing Systems 13 (2000). NPLM์€ ๋‹จ์–ด๋ฅผ ์ž„๋ฒ ๋”ฉํ•˜์—ฌ ๋ฒกํ„ฐ๋กœ ๋ฐ”๊พธ๋Š” ๊ณผ์ •์—์„œ ์‹ ๊ฒฝ๋ง ๊ธฐ๋ฐ˜์˜ ๊ธฐ๋ฒ•์„ ์ œ์‹œํ•˜์—ฌ ํ–ฅํ›„ Word2Vec์œผ๋กœ ๊ฐ€๋Š” ๊ธฐ๋ฐ˜์ด ๋˜์—ˆ๋‹ค๊ณ ํ•œ๋‹ค. ๊ฐ„๋‹จํ•˜๊ฒŒ ํ•™์Šต ๋ฐ์ดํ„ฐ์— ์กด์žฌํ•˜์ง€ ์•Š๋Š” n-gram์ด ํฌํ•จ๋œ ๋ฌธ์žฅ์ด ๋‚˜ํƒ€๋‚  ํ™•๋ฅ ์„ 0์œผ๋กœ ๋งค๊ธด๋‹ค n์„ 5์ด์ƒ์œผ๋กœ ์„ค์ •ํ•˜๊ธฐ ์–ด๋ ต๊ธฐ ๋•Œ๋ฌธ์— ๋ฌธ์žฅ์˜ ์žฅ๊ธฐ ์˜์กด์„ฑ์„ ํฌ์ฐฉํ•ด๋‚ด๊ธฐ ์–ด๋ ต๋‹ค. ๋‹จ์–ด/๋ฌธ์žฅ ๊ฐ„ ์œ ์‚ฌ๋„๋Š” ๊ณ ๋ ค ํ•˜์ง€ ์•Š๋Š”๋‹ค. neural n..
ํ•œ๊ตญ์–ด ๋ฌธ์„œ ์š”์•ฝ ํ‘œํ˜„ ๋…ผ๋ฌธ ์ •๋ฆฌ
ยท
Artificial_Intelligence๐Ÿค–/Natural Language Processing
1) ์ถ”์ถœ์  ์š”์•ฝ(extractive summarization) ์ถ”์ถœ์  ์š”์•ฝ์€ ์›๋ฌธ์—์„œ ์ค‘์š”ํ•œ ํ•ต์‹ฌ ๋ฌธ์žฅ ๋˜๋Š” ๋‹จ์–ด๊ตฌ๋ฅผ ๋ช‡ ๊ฐœ ๋ฝ‘์•„์„œ ์ด๋“ค๋กœ ๊ตฌ์„ฑ๋œ ์š”์•ฝ๋ฌธ์„ ๋งŒ๋“œ๋Š” ๋ฐฉ๋ฒ•์ž…๋‹ˆ๋‹ค. ๊ทธ๋ ‡๊ธฐ ๋•Œ๋ฌธ์— ์ถ”์ถœ์  ์š”์•ฝ์˜ ๊ฒฐ๊ณผ๋กœ ๋‚˜์˜จ ์š”์•ฝ๋ฌธ์˜ ๋ฌธ์žฅ์ด๋‚˜ ๋‹จ์–ด๊ตฌ๋“ค์€ ์ „๋ถ€ ์›๋ฌธ์— ์žˆ๋Š” ๋ฌธ์žฅ๋“ค์ž…๋‹ˆ๋‹ค. ์ถ”์ถœ์  ์š”์•ฝ์˜ ๋Œ€ํ‘œ์ ์ธ ์•Œ๊ณ ๋ฆฌ์ฆ˜์œผ๋กœ ๋จธ์‹  ๋Ÿฌ๋‹ ์•Œ๊ณ ๋ฆฌ์ฆ˜์ธ ํ…์ŠคํŠธ๋žญํฌ(TextRank)๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. 2) ์ถ”์ƒ์  ์š”์•ฝ(abstractive summarization) ์ถ”์ƒ์  ์š”์•ฝ์€ ์›๋ฌธ์— ์—†๋˜ ๋ฌธ์žฅ์ด๋ผ๋„ ํ•ต์‹ฌ ๋ฌธ๋งฅ์„ ๋ฐ˜์˜ํ•œ ์ƒˆ๋กœ์šด ๋ฌธ์žฅ์„ ์ƒ์„ฑํ•ด์„œ ์›๋ฌธ์„ ์š”์•ฝํ•˜๋Š” ๋ฐฉ๋ฒ•์ž…๋‹ˆ๋‹ค. ๋งˆ์น˜ ์‚ฌ๋žŒ์ด ์š”์•ฝํ•˜๋Š” ๊ฒƒ ๊ฐ™์€ ๋ฐฉ์‹์ธ๋ฐ, ๋‹น์—ฐํžˆ ์ถ”์ถœ์  ์š”์•ฝ๋ณด๋‹ค๋Š” ๋‚œ์ด๋„๊ฐ€ ๋†’์Šต๋‹ˆ๋‹ค. ์ด ๋ฐฉ๋ฒ•์€ ์ฃผ๋กœ ์ธ๊ณต ์‹ ๊ฒฝ๋ง์„ ์‚ฌ์šฉํ•˜๋ฉฐ ๋Œ€ํ‘œ์ ์ธ ๋ชจ๋ธ๋กœ seq2seq๊ฐ€ ์žˆ..
DBLP DataSet Processing / ๋Œ€์šฉ๋Ÿ‰ Json ํŒŒ์‹ฑ
ยท
Artificial_Intelligence๐Ÿค–/Natural Language Processing
๊ทธ๋ž˜ํ”„ ์ž„๋ฒ ๋”ฉ์„ ๊ณต๋ถ€ํ•˜๊ธฐ ์œ„ํ•œ DataSet์œผ๋กœ DBLP๋กœ ์ •ํ•˜๊ณ  ์ด๋ฅผ ๊ฐ€์ ธ์™€๋ณด์•˜๋‹ค. https://www.aminer.org/citation AMiner www.aminer.org ์ด ๊ณณ์— ๋“ค์–ด๊ฐ€์„œ ์ด ๋ฐ์ดํ„ฐ๋ฅผ ๊ฐ€์ ธ์™€์„œ ๋‹ค์šด๋กœ๋“œ๋ฅผ ๋ฐ›์•˜๋‹ค. ๊ทธ๋Ÿฐ๋ฐ ๋ฌธ์ œ๋Š” ์ด ๋ฐ์ดํ„ฐ๋ฅผ ๊ฐ€์ ธ์™€์„œ ์ „์ฒ˜๋ฆฌ๋ฅผ ํ•ด์•ผํ•˜๋Š”๋ฐ ์šฉ๋Ÿ‰์ด 16.1GB ์ด๋‹ค.. ์›ฌ๋งŒํ•œ ์—๋””ํ„ฐ๋กœ ์—ด๋ฆฌ์ง€๋„ ์•Š๋Š” ๋ฐ์ดํ„ฐ๋ฅผ ์ฒ˜๋ฆฌํ•ด์•ผํ•ด์„œ ๋ง‰๋ง‰ํ–ˆ์—ˆ๋‹ค. ๊ทธ๋ž˜์„œ ์ƒ๊ฐํ•œ ๊ฒƒ์ด ๋ฐ์ดํ„ฐ๋ฅผ ์šฉ๋Ÿ‰์„ ์ •ํ•ด์„œ ์ž๋ฅด๊ณ , ์ž๋ฅธ ์ฝ”๋“œ๋ฅผ ์ˆ˜์ž‘์—…์œผ๋กœ ์กฐ๊ธˆ๋งŒ ์†๋ด์ฃผ์ž๊ณ  ์ƒ๊ฐํ•˜์˜€๋‹ค. ๋‚ด๊ฐ€ ์‚ฌ์šฉํ•œ ํ”„๋กœ๊ทธ๋žจ์€ GSplit 3 ์ด๋‹ค. ์—ฌ๊ธฐ์„œ ๊ฐ€์ ธ์˜จ DBLP JsonํŒŒ์ผ์„ ๊ฐ€์ ธ์™€์„œ 1GB์”ฉ ๋จผ์ € ์ž˜๋ž๋‹ค. ์ด๋ ‡๊ฒŒ ๋˜๋ฉด, ๋”•์…”๋„ˆ๋ฆฌ๋กœ ์ž๋ฅด๋Š” ๊ฒƒ์ด ์•„๋‹Œ ์šฉ๋Ÿ‰์œผ๋กœ ์ž๋ฅด๊ธฐ์— Json ํ˜•์‹์ด ๊นจ์ง€๊ฒŒ ๋œ๋‹ค. ๋”ฐ๋ผ์„œ, ..
Liky
'Artificial_Intelligence๐Ÿค–/Natural Language Processing' ์นดํ…Œ๊ณ ๋ฆฌ์˜ ๊ธ€ ๋ชฉ๋ก (3 Page)