\ucd5c\uadfc \uba87 \ub144 \uac04 \uc790\uc5f0\uc5b4 \ucc98\ub9ac(Natural Language Processing, NLP) \ubd84\uc57c\ub294 \uae09\uaca9\ud55c \ubc1c\uc804\uc744 \uc774\ub8e8\uc5c8\uc2b5\ub2c8\ub2e4. \uc774\ub294 \uc8fc\ub85c \ub525 \ub7ec\ub2dd \uae30\uc220\uc758 \ubc1c\uc804\uacfc \uadf8\uac83\uc774 \ub300\uaddc\ubaa8 \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\uc5d0 \uc801\uc6a9\ub418\uba74\uc11c \uac00\ub2a5\ud574\uc84c\uc2b5\ub2c8\ub2e4. \uc624\ub298\ub0a0 \uc6b0\ub9ac\ub294 \uc218\ub9ce\uc740 \uc560\ud50c\ub9ac\ucf00\uc774\uc158\uc5d0\uc11c NLP\ub97c \uacbd\ud5d8\ud558\uace0 \uc788\uc73c\uba70, \uadf8 \uc608\ub85c\ub294 \ubc88\uc5ed \uc11c\ube44\uc2a4, \uac1c\uc778 \ube44\uc11c, \uc2a4\ud338 \ud544\ud130\ub9c1, \uac10\uc815 \ubd84\uc11d \ub4f1\uc774 \uc788\uc2b5\ub2c8\ub2e4. \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 \ucf00\ub77c\uc2a4\ub97c \uc774\uc6a9\ud55c \ub525 \ub7ec\ub2dd \uae30\ubc18\uc758 \uc790\uc5f0\uc5b4 \ucc98\ub9ac \uae30\ubc95, \ud2b9\ud788 SimpleRNN\uacfc LSTM(Long Short-Term Memory) \ubaa8\ub378\uc5d0 \ub300\ud574 \uc790\uc138\ud788 \uc54c\uc544\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
\uc790\uc5f0\uc5b4 \ucc98\ub9ac\ub294 \uc778\uac04\uc758 \uc5b8\uc5b4\ub97c \uc774\ud574\ud558\uace0 \ucc98\ub9ac\ud558\ub294 \ucef4\ud4e8\ud130 \uacfc\ud559\uc758 \ud55c \ubd84\uc57c\uc785\ub2c8\ub2e4. NLP\ub294 \ucef4\ud4e8\ud130\uac00 \uc778\uac04\uc758 \uc5b8\uc5b4\ub97c \uc774\ud574\ud558\uace0 \ud574\uc11d\ud560 \uc218 \uc788\ub3c4\ub85d \ud558\ub294 \ub2e4\uc591\ud55c \uae30\uc220\uacfc \uc54c\uace0\ub9ac\uc998\uc744 \ud3ec\ud568\ud569\ub2c8\ub2e4. NLP\uc758 \uc8fc\uc694 \ubaa9\ud45c\ub294 \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\uc5d0\uc11c \uc720\uc6a9\ud55c \uc815\ubcf4\ub97c \ucd94\ucd9c\ud558\uace0, \uc0ac\uc6a9\uc790\uc640\uc758 \uc0c1\ud638\uc791\uc6a9\uc744 \ub354 \uc6d0\ud65c\ud558\uac8c \ub9cc\ub4dc\ub294 \uac83\uc785\ub2c8\ub2e4.<\/p>\n
\ub525 \ub7ec\ub2dd\uc740 \uc778\uacf5 \uc2e0\uacbd\ub9dd\uc744 \ubc1c\uc804\uc2dc\ud0a8 \uae30\uc220\ub85c, \ub370\uc774\ud130\uc758 \uace0\ucc28\uc6d0\uc801\uc778 \ud328\ud134\uc744 \ud559\uc2b5\ud558\ub294 \ub370 \ub6f0\uc5b4\ub09c \uc131\ub2a5\uc744 \ubcf4\uc5ec\uc90d\ub2c8\ub2e4. \ud2b9\ud788 \uc21c\ud658 \uc2e0\uacbd\ub9dd(Recurrent Neural Network, RNN)\uc740 \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\ub97c \ub2e4\ub8e8\ub294 \ub370 \uc801\ud569\ud55c \uad6c\uc870\ub97c \uac00\uc9c0\uace0 \uc788\uc2b5\ub2c8\ub2e4. RNN\uc740 \uac01 \uc2dc\uc810\uc5d0\uc11c \uc774\uc804\uc758 \uc815\ubcf4\ub97c \uae30\uc5b5\ud558\uc5ec \uc2dc\ud000\uc2a4\uc758 \ud750\ub984\uc744 \ubaa8\ub378\ub9c1\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
RNN\uc740 \uae30\uc874\uc758 \ud53c\ub4dc\ud3ec\uc6cc\ub4dc \uc2e0\uacbd\ub9dd\uacfc \ub2e4\ub974\uac8c \uc815\ubcf4\uac00 \uc774\uc804 \uc0c1\ud0dc\ub85c \ud750\ub97c \uc218 \uc788\ub294 \uad6c\uc870\ub97c \uac00\uc9c0\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \uad6c\uc870 \ub355\ubd84\uc5d0 RNN\uc740 \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\uc758 \uc2dc\uac04\uc801 \uc758\uc874\uc131\uc744 \ubaa8\ub378\ub9c1\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ud558\uc9c0\ub9cc RNN\uc740 \uae34 \uc2dc\ud000\uc2a4\ub97c \ub2e4\ub8f0 \ub54c ‘\uae30\uc5b5’\uc758 \ubb38\uc81c\uac00 \ubc1c\uc0dd\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc989, \uc774\uc804\uc758 \uc815\ubcf4\uac00 \uc2dc\uac04\uc774 \uc9c0\ub0a8\uc5d0 \ub530\ub77c \uc0ac\ub77c\uc9c8 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
SimpleRNN\uc740 \uac00\uc7a5 \uae30\ubcf8\uc801\uc778 \ud615\ud0dc\uc758 RNN\uc73c\ub85c, \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\uc5d0\uc11c \ub2e8\uae30 \uae30\uc5b5\uc744 \ub2e4\ub8e8\ub294 \ub370 \uc720\uc6a9\ud569\ub2c8\ub2e4. \ud558\uc9c0\ub9cc \uae34 \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\ub97c \ucc98\ub9ac\ud558\ub294 \ub370\ub294 \ud55c\uacc4\uac00 \uc788\uc2b5\ub2c8\ub2e4. SimpleRNN\uc758 \uc218\uc2dd\uc740 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4:<\/p>\n
h_t = tanh(W_hh * h_{t-1} + W_xh * x_t + b_h)<\/code><\/pre>\n\uc5ec\uae30\uc11c h_t<\/code>\ub294 \uc2dc\uac04 t<\/code>\uc5d0\uc11c\uc758 \uc740\ub2c9 \uc0c1\ud0dc, x_t<\/code>\ub294 \uc785\ub825, W_hh<\/code>, W_xh<\/code>\ub294 \uac00\uc911\uce58, b_h<\/code>\ub294 \ud3b8\ud5a5\uc744 \ub098\ud0c0\ub0c5\ub2c8\ub2e4.<\/p>\n2.3 SimpleRNN\uc758 \ud55c\uacc4<\/h3>\n
SimpleRNN\uc740 \ub2e8\uae30\uc801\uc778 \uc758\uc874\uc131\uc740 \uc798 \ucc98\ub9ac\ud558\uc9c0\ub9cc, \uc7a5\uae30\uc801\uc778 \uc758\uc874\uc131\uc744 \ucc98\ub9ac\ud558\ub294 \ub370 \uc5b4\ub824\uc6c0\uc744 \uacaa\uc2b5\ub2c8\ub2e4. \uc65c\ub0d0\ud558\uba74, \uc5ed\uc804\ud30c \uc911\uc5d0 gradient vanishing \ubb38\uc81c\uac00 \ubc1c\uc0dd\ud558\uc5ec, \uc6d0\uac70\ub9ac\uc758 \uc785\ub825\uc5d0 \ub300\ud55c \uc601\ud5a5\ub825\uc774 \uc0ac\ub77c\uc9c0\uae30 \ub54c\ubb38\uc785\ub2c8\ub2e4.<\/p>\n
3. LSTM(Long Short-Term Memory)<\/h2>\n
LSTM\uc740 RNN\uc758 \ubc1c\uc804\ud615\uc73c\ub85c, \uc7a5\uae30\uc801\uc778 \uc758\uc874\uc131\uc744 \ubaa8\ub378\ub9c1\ud558\uae30 \uc704\ud574 \uc124\uacc4\ub418\uc5c8\uc2b5\ub2c8\ub2e4. LSTM\uc740 \uc140 \uc0c1\ud0dc(cell state)\uc640 \uac8c\uc774\ud2b8(gate) \uba54\ucee4\ub2c8\uc998\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc815\ubcf4\uc758 \ud750\ub984\uc744 \uc870\uc808\ud569\ub2c8\ub2e4. \uc774 \uad6c\uc870 \ub355\ubd84\uc5d0 LSTM\uc740 \uae34 \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\uc5d0\uc11c\ub3c4 \ud6a8\uacfc\uc801\uc73c\ub85c \uc815\ubcf4\ub97c \uae30\uc5b5\ud558\uace0 \uc78a\uc5b4\ubc84\ub9b4 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
3.1 LSTM\uc758 \uad6c\uc870<\/h3>\n
LSTM\uc740 \uae30\ubcf8\uc801\uc73c\ub85c \uc140 \uc0c1\ud0dc, \uc785\ub825 \uac8c\uc774\ud2b8, \ucd9c\ub825 \uac8c\uc774\ud2b8, \ub9dd\uac01 \uac8c\uc774\ud2b8\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4. \uac01 \uac8c\uc774\ud2b8\ub294 \ud2b9\uc815 \uc815\ubcf4\ub97c \ud1b5\uacfc\uc2dc\ud0a4\uac70\ub098 \ucc28\ub2e8\ud558\uc5ec \uc815\ubcf4\uc758 \ud750\ub984\uc744 \uc870\uc808\ud569\ub2c8\ub2e4.<\/p>\n
\n- \uc785\ub825 \uac8c\uc774\ud2b8:<\/strong> \uc0c8\ub85c\uc6b4 \uc815\ubcf4\ub97c \uc5bc\ub9c8\ub098 \ubc1b\uc544\ub4e4\uc77c\uc9c0 \uacb0\uc815\ud569\ub2c8\ub2e4.<\/li>\n
- \ub9dd\uac01 \uac8c\uc774\ud2b8:<\/strong> \uc140 \uc0c1\ud0dc\uc5d0\uc11c \uc5b4\ub5a4 \uc815\ubcf4\ub97c \uc78a\uc5b4\ubc84\ub9b4\uc9c0 \uacb0\uc815\ud569\ub2c8\ub2e4.<\/li>\n
- \ucd9c\ub825 \uac8c\uc774\ud2b8:<\/strong> \ucd5c\uc885 \ucd9c\ub825\uc744 \uacb0\uc815\ud569\ub2c8\ub2e4.<\/li>\n<\/ul>\n
3.2 LSTM\uc758 \uc218\uc2dd<\/h3>\n
LSTM\uc758 \uc218\uc2dd\uc740 \ub2e4\uc74c\uacfc \uac19\uc774 \ud45c\ud604\ub429\ub2c8\ub2e4:<\/p>\n
f_t = \u03c3(W_f * [h_{t-1}, x_t] + b_f) \/\/ \ub9dd\uac01 \uac8c\uc774\ud2b8\ni_t = \u03c3(W_i * [h_{t-1}, x_t] + b_i) \/\/ \uc785\ub825 \uac8c\uc774\ud2b8\no_t = \u03c3(W_o * [h_{t-1}, x_t] + b_o) \/\/ \ucd9c\ub825 \uac8c\uc774\ud2b8\n\n\\hat{C}_t = tanh(W_C * [h_{t-1}, x_t] + b_C) \/\/ \ud6c4\ubcf4 \uc140 \uc0c1\ud0dc\n\nC_t = f_t * C_{t-1} + i_t * \\hat{C}_t \/\/ \uc140 \uc0c1\ud0dc \uc5c5\ub370\uc774\ud2b8\nh_t = o_t * tanh(C_t) \/\/ \ucd5c\uc885 \ucd9c\ub825\n<\/code><\/pre>\n\uc5ec\uae30\uc11c f_t<\/code>, i_t<\/code>, o_t<\/code>\ub294 \uac01\uac01 \ub9dd\uac01, \uc785\ub825, \ucd9c\ub825 \uac8c\uc774\ud2b8\ub97c \ub098\ud0c0\ub0b4\uace0, C_t<\/code>\ub294 \uc140 \uc0c1\ud0dc, h_t<\/code>\ub294 \ucd9c\ub825 \uc0c1\ud0dc\uc785\ub2c8\ub2e4.<\/p>\n3.3 LSTM\uc758 \uc7a5\uc810<\/h3>\n
LSTM\uc758 \uac00\uc7a5 \ud070 \uc7a5\uc810\uc740 \uae34 \uc2dc\ud000\uc2a4\uc5d0\uc11c\ub3c4 \uc815\ubcf4\uc758 \uc190\uc2e4\uc774 \uc801\ub2e4\ub294 \uc810\uc785\ub2c8\ub2e4. \uc774\ub294 \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc640 \uac19\uc740 \ubd84\uc57c\uc5d0\uc11c \uc911\uc694\ud55c \ud2b9\uc9d5\uc73c\ub85c \uc791\uc6a9\ud558\uba70, \uae30\uacc4 \ubc88\uc5ed, \uac10\uc815 \ubd84\uc11d \ub4f1 \uc5ec\ub7ec \uc751\uc6a9 \ubd84\uc57c\uc5d0\uc11c \uc6b0\uc218\ud55c \uc131\ub2a5\uc744 \ubcf4\uc5ec\uc90d\ub2c8\ub2e4.<\/p>\n
4. \ucf00\ub77c\uc2a4(Keras)\ub97c \uc774\uc6a9\ud55c \ubaa8\ub378 \uad6c\ud604<\/h2>\n
\ucf00\ub77c\uc2a4\ub294 \ud30c\uc774\uc36c\uc73c\ub85c \uc791\uc131\ub41c \uace0\uc218\uc900\uc758 \ub525 \ub7ec\ub2dd API\ub85c, TensorFlow, Theano \ub4f1\uc758 \ubc31\uc5d4\ub4dc\uc5d0\uc11c \uc791\ub3d9\ud569\ub2c8\ub2e4. \ucf00\ub77c\uc2a4\ub97c \uc0ac\uc6a9\ud558\uba74 \ubcf5\uc7a1\ud55c \ub525 \ub7ec\ub2dd \ubaa8\ub378\uc744 \ube44\uad50\uc801 \uc27d\uac8c \uad6c\ucd95\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ubc88 \uc139\uc158\uc5d0\uc11c\ub294 SimpleRNN\uacfc LSTM \ubaa8\ub378\uc744 \ucf00\ub77c\uc2a4\ub97c \uc0ac\uc6a9\ud558\uc5ec \uad6c\ud604\ud558\ub294 \ubc29\ubc95\uc744 \uc54c\uc544\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
4.1 \ud658\uacbd \uc124\uc815<\/h3>\n
\ucf00\ub77c\uc2a4\ub97c \uc0ac\uc6a9\ud558\uae30 \uc704\ud574 \uc6b0\uc120 \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc124\uce58\ud574\uc57c \ud569\ub2c8\ub2e4. \uc544\ub798\uc758 \uba85\ub839\uc5b4\ub97c \ud1b5\ud574 \ucf00\ub77c\uc2a4\ub97c \uc124\uce58\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\n
pip install keras tensorflow<\/code><\/pre>\n4.2 \ub370\uc774\ud130 \uc804\ucc98\ub9ac<\/h3>\n
\uc790\uc5f0\uc5b4 \ub370\uc774\ud130\ub294 \ubaa8\ub378\uc5d0 \uc785\ub825\ud558\uae30 \uc804\uc5d0 \uc801\uc808\ud55c \ud615\ud0dc\ub85c \uac00\uacf5\ud574\uc57c \ud569\ub2c8\ub2e4. \uc77c\ubc18\uc801\uc73c\ub85c \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\ub97c \uc815\uc218 \uc778\ucf54\ub529 \ub610\ub294 \uc6d0-\ud56b \uc778\ucf54\ub529\uc744 \ud1b5\ud574 \uc218\uce58\ud615 \ub370\uc774\ud130\ub85c \ubcc0\ud658\ud569\ub2c8\ub2e4. \uc544\ub798\ub294 \ub370\uc774\ud130\ub97c \uc804\ucc98\ub9ac\ud558\ub294 \ubc29\ubc95\uc758 \uc608\uc2dc\uc785\ub2c8\ub2e4:<\/p>\n
\nfrom keras.preprocessing.text import Tokenizer\nfrom keras.preprocessing.sequence import pad_sequences\n\n# \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\ntexts = [\"\ub098\ub294 \ubc25\uc744 \uba39\uc5c8\ub2e4\", \"\uc624\ub298 \ub0a0\uc528\uac00 \uc88b\ub2e4\", \"\ubb34\uc2a8 \uc601\ud654\ub97c \ubcf4\uace0 \uc2f6\uc5b4\"]\n\n# \ud1a0\ud06c\ub098\uc774\uc800 \uc0dd\uc131 \ubc0f \ud14d\uc2a4\ud2b8 \uc2dc\ud000\uc2a4 \ubcc0\ud658\ntokenizer = Tokenizer()\ntokenizer.fit_on_texts(texts)\nsequences = tokenizer.texts_to_sequences(texts)\n\n# \uc2dc\ud000\uc2a4 \ud328\ub529\nmaxlen = 5 # \ucd5c\ub300 \uae38\uc774 \uc124\uc815\ndata = pad_sequences(sequences, maxlen=maxlen)\n<\/code><\/pre>\n4.3 SimpleRNN \ubaa8\ub378 \uad6c\ucd95<\/h3>\n
\uc774\uc81c SimpleRNN \ubaa8\ub378\uc744 \uad6c\ucd95\ud574 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc544\ub798\uc758 \ucf54\ub4dc\ub97c \ud1b5\ud574 \uac04\ub2e8\ud55c SimpleRNN \ubaa8\ub378\uc744 \uc218\ub9bd\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\n
\nfrom keras.models import Sequential\nfrom keras.layers import SimpleRNN, Dense, Embedding\n\n# \ubaa8\ub378 \uc0dd\uc131\nmodel = Sequential()\nmodel.add(Embedding(input_dim=len(tokenizer.word_index)+1, output_dim=8, input_length=maxlen))\nmodel.add(SimpleRNN(8))\nmodel.add(Dense(1, activation='sigmoid'))\n\n# \ucef4\ud30c\uc77c\nmodel.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])\n\n# \ubaa8\ub378 \uc694\uc57d\nmodel.summary()\n<\/code><\/pre>\n4.4 LSTM \ubaa8\ub378 \uad6c\ucd95<\/h3>\n
\uc774\ubc88\uc5d0\ub294 LSTM \ubaa8\ub378\uc744 \uad6c\ucd95\ud574 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc544\ub798\uc5d0 LSTM \ubaa8\ub378\uc758 \uc608\uc2dc \ucf54\ub4dc\ub97c \uc81c\uacf5\ud569\ub2c8\ub2e4:<\/p>\n
\nfrom keras.layers import LSTM\n\n# \ubaa8\ub378 \uc0dd\uc131\nmodel = Sequential()\nmodel.add(Embedding(input_dim=len(tokenizer.word_index)+1, output_dim=8, input_length=maxlen))\nmodel.add(LSTM(8))\nmodel.add(Dense(1, activation='sigmoid'))\n\n# \ucef4\ud30c\uc77c\nmodel.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])\n\n# \ubaa8\ub378 \uc694\uc57d\nmodel.summary()\n<\/code><\/pre>\n4.5 \ubaa8\ub378 \ud6c8\ub828<\/h3>\n
\ubaa8\ub378\uc774 \uad6c\ucd95\ub418\uc5c8\uc73c\uba74 \uc774\uc81c \ub370\uc774\ud130\ub97c \uc0ac\uc6a9\ud574 \ud6c8\ub828\uc2dc\ud0ac \ucc28\ub840\uc785\ub2c8\ub2e4. \uc801\uc808\ud55c \ub370\uc774\ud130\uc14b\uc744 \uac00\uc9c0\uace0 \uc55e\uc11c \ub9cc\ub4e4\uc5c8\ub358 \ubaa8\ub378\uc744 \ud559\uc2b5\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\n# \uc608\uc2dc \ub370\uc774\ud130 (X\ub294 \uc785\ub825, y\ub294 \ub77c\ubca8)\nX = data\ny = [1, 0, 1] # \ub77c\ubca8 \uc608\uc2dc\n\nmodel.fit(X, y, epochs=10, batch_size=2)\n<\/code><\/pre>\n5. \uacb0\ub860<\/h2>\n
\uc774\ubc88 \uac15\uc88c\uc5d0\uc11c\ub294 \ub525 \ub7ec\ub2dd\uc744 \uc774\uc6a9\ud55c \uc790\uc5f0\uc5b4 \ucc98\ub9ac \ubd84\uc57c\uc5d0\uc11c\uc758 SimpleRNN\uacfc LSTM\uc5d0 \ub300\ud574 \ub2e4\ub8e8\uc5c8\uc2b5\ub2c8\ub2e4. RNN\uc740 \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\ub97c \ub2e4\ub8e8\ub294 \ub370 \ud544\uc218\uc801\uc778 \ubaa8\ub378\uc774\uc9c0\ub9cc, \uc7a5\uae30\uc801\uc778 \uc758\uc874\uc131\uacfc \uad00\ub828\ub41c \ubb38\uc81c\ub85c \uc778\ud574 LSTM\uc774 \ubc1c\uc804\ud558\uac8c \ub418\uc5c8\uc2b5\ub2c8\ub2e4. LSTM\uc740 \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc5d0\uc11c \ud6a8\uacfc\uc801\uc778 \uc131\ub2a5\uc744 \ubc1c\ud718\ud558\uba70, \ucf00\ub77c\uc2a4\ub97c \ud1b5\ud574 \uc27d\uac8c \uad6c\ud604\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\uc774\uc640 \uac19\uc740 \ubc29\ubc95\uc73c\ub85c \ub2e4\uc591\ud55c NLP \uacfc\uc81c\ub97c \ud574\uacb0\ud560 \uc218 \uc788\uc73c\uba70, \uc55e\uc73c\ub85c\ub3c4 \ub525 \ub7ec\ub2dd \uae30\uc220\uc774 \uc9c0\uc18d\uc801\uc73c\ub85c \ubc1c\uc804\ud558\uc5ec \ub354 \ub098\uc740 \uc790\uc5f0\uc5b4 \ucc98\ub9ac \ubaa8\ub378\uc774 \ub4f1\uc7a5\ud560 \uac83\uc785\ub2c8\ub2e4.<\/p>\n
6. \ucc38\uace0 \uc790\ub8cc<\/h2>\n