[CS234 + NLP] Lecture 3. Word Embadding(2)
ยท
๐Ÿฆ„AI/NLP
[NLP] 1. Word Embadding (1)
ยท
๐Ÿฆ„AI/NLP
Word Vectors์–ด๋–ค ๋ฌธ์žฅ์ด ์ฃผ์–ด์กŒ์„๋•Œ, ๋ฌธ์žฅ์˜ ์–ด๋–ค ๋ถ€๋ถ„์ด ๋ฌธ์žฅ์˜ ์˜๋ฏธ๋ฅผ ๋” ์ค‘์š”ํ•˜๊ฒŒ ํ‘œํ˜„ํ•˜๋Š”์ง€๋ฅผ ์–ด๋–ป๊ฒŒ ์•Œ ์ˆ˜ ์žˆ์„๊นŒ?์‚ฌ๋žŒ์˜ ์–ธ์–ด๋ผ๋Š” ๊ฒƒ์€ ๋‹ค์–‘ํ•œ ์˜๋ฏธ์™€ ๋‰˜์•™์Šค๋ฅผ ๊ฐ€์ง€๊ณ  ์žˆ๊ธฐ ๋•Œ๋ฌธ์—, ์–ธ์–ด์˜ ์ •๋ณด๋ฅผ ์ž˜ ํฌํ•จํ•˜๊ฒŒ ํ‘œํ˜„ํ•˜๊ณ , ์ด๋ฅผ ๋‹ค๋ฃจ๋Š”๊ฒƒ์€ ์ •๋ง ์–ด๋ ค์šด ์ผ์ด๋‹ค.one hot vectorword๋ฅผ ๋‹ค๋ฃจ๋Š” ๊ฐ€์žฅ ์‰ฝ๊ณ  ๊ธฐ๋ณธ์ ์ธ ๋ฐฉ๋ฒ•์€ ๋‹จ์–ด๋ฅผ ์„œ๋กœ ์˜์กดํ•˜์ง€ ์•Š๋Š” ๊ฐœ๋ณ„์˜ ๊ฐœ์ฒด๋กœ ์ƒ๊ฐํ•˜๋Š” ๊ฒƒ์ด๋‹ค.์ฆ‰, ๋‹จ์–ด๋ฅผ one hot vector๋ฅผ ํ†ตํ•ด์„œ ๋‚˜ํƒ€๋‚ผ ์ˆ˜ ์žˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด์„œ finite set {coffee, cafe, tea}์˜ ๊ฐœ๋ณ„ ๋‹จ์–ด๋ฅผ ๋‹ค์Œ์ฒ˜๋Ÿผ 1-hot vector๋กœ ํ‘œํ˜„ํ•œ๋‹ค.์ „ํ†ต์ ๋”˜ NLP์—์„œ ๋‹จ์–ด๋ฅผ ์ด์™€ ๊ฐ™์€ ํ˜•ํƒœ๋กœ ํ‘œํ˜„ํ•˜์—ฌ ์‚ฌ์šฉํ•œ๋‹ค. ์ด๋ ‡๊ฒŒ ํ‘œํ˜„ํ•˜๊ฒŒ ๋˜๋ฉด, ๋‘ ๋‹จ์–ด ์‚ฌ์ด์˜ L1, L2 distan..