\uc790\uc5f0\uc5b4 \ucc98\ub9ac\ub294 \ucef4\ud4e8\ud130\uac00 \uc778\uac04\uc758 \uc5b8\uc5b4\ub97c \uc774\ud574\ud558\uace0 \ud574\uc11d\ud560 \uc218 \uc788\ub3c4\ub85d \ud558\ub294 \uae30\uc220\uc785\ub2c8\ub2e4. \uc774 \uae30\uc220\uc740 \ud14d\uc2a4\ud2b8 \ubd84\uc11d, \ubc88\uc5ed, \uac10\uc815 \ubd84\uc11d \ub4f1 \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0\uc11c \uc0ac\uc6a9\ub418\uba70, \ucd5c\uadfc \uba87 \ub144\uac04 \ub525 \ub7ec\ub2dd\uc758 \ubc1c\uc804\uc73c\ub85c \ub354\uc6b1 \ube60\ub974\uace0 \uc815\ud655\ud55c \ucc98\ub9ac \ub2a5\ub825\uc744 \ubcf4\uc5ec\uc8fc\uace0 \uc788\uc2b5\ub2c8\ub2e4. \ud2b9\ud788, \uc758\ub3c4 \ubd84\ub958(Intent Classification)\ub294 \uc0ac\uc6a9\uc790\uc758 \uc785\ub825\uc5d0\uc11c \ud2b9\uc815\ud55c \uc758\ub3c4\ub97c \ucd94\ucd9c\ud558\ub294 \uacfc\uc815\uc73c\ub85c, \ucc57\ubd07\uc774\ub098 \uace0\uac1d \uc9c0\uc6d0 \uc2dc\uc2a4\ud15c, \uc74c\uc131 \uc778\uc2dd \uc2dc\uc2a4\ud15c \ub4f1\uc5d0 \ud544\uc218\uc801\uc778 \uae30\ub2a5\uc785\ub2c8\ub2e4.<\/p>\n
\uc758\ub3c4 \ubd84\ub958\ub294 \uc8fc\uc5b4\uc9c4 \ubb38\uc7a5\uc774\ub098 \uc9c8\ubb38\uc5d0\uc11c \uc0ac\uc6a9\uc790\uc758 \uc758\ub3c4\ub97c \ud30c\uc545\ud558\ub294 \uc791\uc5c5\uc785\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, “\ub0b4\uc77c \ub0a0\uc528 \uc5b4\ub54c?”\ub77c\ub294 \uc9c8\ubb38\uc774 \uc788\uc744 \ub54c, \uae30\ubcf8\uc801\uc73c\ub85c \uc0ac\uc6a9\uc790\ub294 \ub0a0\uc528 \uc815\ubcf4\ub97c \uc694\uccad\ud558\ub294 \uc758\ub3c4\ub97c \uac00\uc9c0\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \uc758\ub3c4\ub97c \uc815\ud655\ud558\uac8c \uc2dd\ubcc4\ud558\uc5ec \uc801\uc808\ud55c \uc751\ub2f5\uc744 \uc81c\uacf5\ud558\ub294 \uac83\uc774 \uc758\ub3c4 \ubd84\ub958\uc758 \ubaa9\uc801\uc785\ub2c8\ub2e4.<\/p>\n
\uc804\ud1b5\uc801\uc778 \uc790\uc5f0\uc5b4 \ucc98\ub9ac \uae30\ubc95\uc740 \uc8fc\ub85c \uaddc\uce59 \uae30\ubc18 \uc2dc\uc2a4\ud15c\uc774\ub098 \uba38\uc2e0 \ub7ec\ub2dd \uae30\uc220\uc5d0 \uc758\uc874\ud588\uc2b5\ub2c8\ub2e4. \uadf8\ub7ec\ub098 \ub525 \ub7ec\ub2dd\uc758 \ub4f1\uc7a5\uc73c\ub85c \uc778\ud574 \ub300\ub7c9\uc758 \ub370\uc774\ud130\uc5d0\uc11c \ud328\ud134\uc744 \ud559\uc2b5\ud558\uace0, \ub354\uc6b1 \ubcf5\uc7a1\ud55c \ubaa8\ub378\uc744 \ud1b5\ud574 \uc790\uc5f0\uc5b4\uc758 \ub258\uc559\uc2a4\uc640 \ub9e5\ub77d\uc744 \uc774\ud574\ud560 \uc218 \uc788\uac8c \ub418\uc5c8\uc2b5\ub2c8\ub2e4. \ub525 \ub7ec\ub2dd\uc740 \ud2b9\ud788 \ub2e4\uc74c\uacfc \uac19\uc740 \uacbd\uc6b0\uc5d0 \uac15\ub825\ud55c \uc131\ub2a5\uc744 \ubc1c\ud718\ud569\ub2c8\ub2e4:<\/p>\n
\uc6cc\ub4dc \uc784\ubca0\ub529\uc740 \ub2e8\uc5b4\ub97c \ubca1\ud130 \uacf5\uac04\uc5d0 \ub9e4\ud551\ud558\uc5ec \uac01 \ub2e8\uc5b4 \uac04\uc758 \uc758\ubbf8\uc801 \uc720\uc0ac\uc131\uc744 \ubc18\uc601\ud558\ub294 \ubc29\ubc95\uc785\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, “king”\uacfc “queen”\uc740 \uc758\ubbf8\uc801\uc73c\ub85c \uc720\uc0ac\ud558\ubbc0\ub85c, \uc774\ub7ec\ud55c \ub2e8\uc5b4\ub4e4\uc740 \ubca1\ud130 \uacf5\uac04\uc5d0\uc11c \uac00\uae4c\uc6b4 \uc704\uce58\uc5d0 \ubc30\uce58\ub429\ub2c8\ub2e4. \uc0ac\uc804 \ud6c8\ub828\ub41c \uc6cc\ub4dc \uc784\ubca0\ub529\uc744 \uc0ac\uc6a9\ud558\uba74 \ub2e4\uc74c\uacfc \uac19\uc740 \uc7a5\uc810\uc744 \uc5bb\uc744 \uc218 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\n
\uc0ac\uc804 \ud6c8\ub828\ub41c \uc6cc\ub4dc \uc784\ubca0\ub529\uc758 \ub300\ud45c\uc801\uc778 \uc608\ub85c\ub294 \ub2e4\uc74c\uacfc \uac19\uc740 \ubaa8\ub378\uc774 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\n
Word2Vec\ub294 \uad6c\uae00\uc5d0\uc11c \uac1c\ubc1c\ud55c \ubaa8\ub378\ub85c, Continuous Bag of Words (CBOW)\uc640 Skip-gram \ub450 \uac00\uc9c0 \uc544\ud0a4\ud14d\ucc98\ub97c \uc0ac\uc6a9\ud558\uc5ec \ub2e8\uc5b4\uc758 \uc784\ubca0\ub529\uc744 \ud559\uc2b5\ud569\ub2c8\ub2e4. CBOW\ub294 \uc8fc\ubcc0 \ub2e8\uc5b4\ub4e4\uc744 \uc774\uc6a9\ud574 \uc911\uc2ec \ub2e8\uc5b4\ub97c \uc608\uce21\ud558\ub294 \ubc29\uc2dd\uc774\uace0, Skip-gram\uc740 \uc911\uc2ec \ub2e8\uc5b4\ub97c \uc774\uc6a9\ud574 \uc8fc\ubcc0 \ub2e8\uc5b4\ub4e4\uc744 \uc608\uce21\ud569\ub2c8\ub2e4.<\/p>\n
GloVe(Global Vectors for Word Representation)\ub294 \ud398\uc774\uc2a4\ubd81\uc5d0\uc11c \uac1c\ubc1c\ud55c \ubaa8\ub378\ub85c, \uc804\uc5ed\uc801\uc778 \ud1b5\uacc4 \uc815\ubcf4\ub97c \ud1b5\ud574 \ub2e8\uc5b4 \uac04\uc758 \uad00\uacc4\ub97c \ud30c\uc545\ud569\ub2c8\ub2e4. GloVe\ub294 \ub2e8\uc5b4 \uac04\uc758 \ub3d9\uc2dc \ubc1c\uc0dd \ud655\ub960\uc744 \uae30\ubc18\uc73c\ub85c \ud558\uc5ec \uc758\ubbf8\uc801\uc73c\ub85c \uc720\uc0ac\ud55c \ub2e8\uc5b4\ub4e4\uc744 \ubca1\ud130 \uacf5\uac04\uc5d0\uc11c \uac00\uae5d\uac8c \ubc30\uce58\ud569\ub2c8\ub2e4.<\/p>\n
FastText\ub294 \ud398\uc774\uc2a4\ubd81\uc5d0\uc11c \uac1c\ubc1c\ud55c \ubaa8\ub378\ub85c, \ub2e8\uc5b4\ub97c \ubb38\uc790 \ub2e8\uc704\ub85c \ub098\ub204\uc5b4 \ud559\uc2b5\ud558\ub294 \ubc29\ubc95\uc785\ub2c8\ub2e4. \uc774\ub85c \uc778\ud574 \ub3d9\uc77c\ud55c \uc758\ubbf8\ub97c \uac00\uc9c4 \ub2e8\uc5b4\ub4e4(\uc608: “playing”\uacfc “play”)\uc758 \ubca1\ud130\uac00 \ube44\uc2b7\ud558\uac8c \ud615\uc131\ub418\uc5b4, OOV(Out-Of-Vocabulary) \ubb38\uc81c\ub97c \ud6a8\uacfc\uc801\uc73c\ub85c \ud574\uacb0\ud569\ub2c8\ub2e4.<\/p>\n
\uc774\uc81c \uc758\ub3c4 \ubd84\ub958\ub97c \uc704\ud55c \ub525 \ub7ec\ub2dd \ubaa8\ub378\uc744 \uad6c\ucd95\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc740 \ud06c\uac8c \ub370\uc774\ud130 \uc218\uc9d1 \ubc0f \uc804\ucc98\ub9ac, \ubaa8\ub378 \uc124\uacc4, \ud6c8\ub828 \ubc0f \ud3c9\uac00\uc758 \ub2e8\uacc4\ub85c \ub098\ub269\ub2c8\ub2e4.<\/p>\n
\uc758\ub3c4 \ubd84\ub958 \ubaa8\ub378\uc744 \uad6c\ucd95\ud558\uae30 \uc704\ud574\uc11c\ub294 \uc758\ub3c4\uc640 \uad00\ub828\ub41c \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\uac00 \ud544\uc694\ud569\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ub370\uc774\ud130\ub294 \uacf5\uac1c \ub370\uc774\ud130\uc14b\uc744 \uc0ac\uc6a9\ud560 \uc218\ub3c4 \uc788\uace0, \uc6f9 \uc2a4\ud06c\ub798\ud551 \ub4f1\uc744 \ud1b5\ud574 \uc218\uc9d1\ud560 \uc218\ub3c4 \uc788\uc2b5\ub2c8\ub2e4. \ub370\uc774\ud130 \uc218\uc9d1 \ud6c4\uc5d0\ub294 \ub2e4\uc74c\uacfc \uac19\uc740 \uc804\ucc98\ub9ac \uacfc\uc815\uc744 \uac70\uccd0\uc57c \ud569\ub2c8\ub2e4:<\/p>\n
\uc758\ub3c4 \ubd84\ub958\ub97c \uc704\ud55c \ubaa8\ub378\uc740 \uc8fc\ub85c \uc21c\ud658 \uc2e0\uacbd\ub9dd(RNN)\uc774\ub098 \uc7a5\uae30 \ub2e8\uae30 \uba54\ubaa8\ub9ac(LSTM) \ub124\ud2b8\uc6cc\ud06c\ub97c \uc774\uc6a9\ud558\uc5ec \uc124\uacc4\ub429\ub2c8\ub2e4. \ub2e4\uc74c\uc740 \uac04\ub2e8\ud55c LSTM \ubaa8\ub378\uc758 \uc608\uc2dc\uc785\ub2c8\ub2e4:<\/p>\n
import tensorflow as tf\nfrom tensorflow.keras.preprocessing.text import Tokenizer\nfrom tensorflow.keras.preprocessing.sequence import pad_sequences\nfrom tensorflow.keras.models import Sequential\nfrom tensorflow.keras.layers import Embedding, LSTM, Dense\n\n# \ub370\uc774\ud130 \uc900\ube44\nsentences = ['\ubb38\uc7a51', '\ubb38\uc7a52', ...] # \uc758\ub3c4\ubd84\ub958\ub97c \uc704\ud55c \ubb38\uc7a5 \ub9ac\uc2a4\ud2b8\nlabels = [0, 1, ...] # \uac01 \ubb38\uc7a5\uc5d0 \ub300\ud55c \ub77c\ubca8\n\n# \ud1a0\ud070\ud654\ntokenizer = Tokenizer()\ntokenizer.fit_on_texts(sentences)\nsequences = tokenizer.texts_to_sequences(sentences)\n\n# \ud328\ub529\nmax_length = max(len(seq) for seq in sequences)\npadded_sequences = pad_sequences(sequences, maxlen=max_length, padding='post')\n\n# \ubaa8\ub378 \uad6c\ucd95\nmodel = Sequential()\nmodel.add(Embedding(input_dim=len(tokenizer.word_index) + 1, output_dim=100, input_length=max_length))\nmodel.add(LSTM(128))\nmodel.add(Dense(1, activation='sigmoid'))\n\nmodel.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])\n<\/code><\/pre>\n5.3 \ubaa8\ub378 \ud6c8\ub828<\/h3>\n
\ud6c8\ub828 \ub370\uc774\ud130\uc640 \uac80\uc99d \ub370\uc774\ud130\ub85c \ubaa8\ub378\uc744 \ud6c8\ub828\uc2dc\ud0b5\ub2c8\ub2e4. \ud6c8\ub828 \uacfc\uc815\uc740 \uc5ec\ub7ec epoch\uc744 \uac70\uccd0 \ubaa8\ub378\uc758 \uc131\ub2a5\uc744 \uc9c0\uc18d\uc801\uc73c\ub85c \uac1c\uc120\ud558\ub294 \uacfc\uc815\uc785\ub2c8\ub2e4.<\/p>\n
# \ubaa8\ub378 \ud6c8\ub828\nmodel.fit(padded_sequences, labels, epochs=10, validation_split=0.2)\n<\/code><\/pre>\n5.4 \ubaa8\ub378 \ud3c9\uac00 \ubc0f \uc608\uce21<\/h3>\n
\ud6c8\ub828\uc774 \uc644\ub8cc\ub41c \ud6c4, \uac80\uc99d \ub370\uc774\ud130\ub85c \ubaa8\ub378\uc744 \ud3c9\uac00\ud558\uace0 \uc2e4\uc81c \ub370\uc774\ud130\ub97c \uc785\ub825\ud558\uc5ec \uc608\uce21\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
# \ubaa8\ub378 \ud3c9\uac00\nloss, accuracy = model.evaluate(validation_data)\n\n# \uc608\uce21\npredictions = model.predict(new_data)\n<\/code><\/pre>\n6. \uc2e4\uc6a9\uc801\uc778 \uc751\uc6a9<\/h2>\n
\uc758\ub3c4 \ubd84\ub958 \ubaa8\ub378\uc740 \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0\uc11c \uc2e4\uc6a9\uc801\uc73c\ub85c \ud65c\uc6a9\ub420 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ud2b9\ud788 \uace0\uac1d \uc11c\ube44\uc2a4\uc5d0\uc11c\uc758 \ucc57\ubd07, \uc74c\uc131 \uae30\ubc18 \ube44\uc11c, \uc2a4\ud338 \uba54\uc77c \ud544\ud130\ub9c1, \ub9ac\ubdf0 \ubd84\uc11d \ub4f1\uc5d0\uc11c \ud544\uc218\uc801\uc778 \uc5ed\ud560\uc744 \ud569\ub2c8\ub2e4.<\/p>\n
6.1 \ucc57\ubd07<\/h3>\n
\ucc57\ubd07\uc740 \uace0\uac1d\uc758 \uc9c8\ubb38\uc5d0 \ub300\ud55c \uc790\ub3d9\ud654\ub41c \ub2f5\ubcc0\uc744 \uc81c\uacf5\ud558\ub294 \ub3c4\uad6c\ub85c, \uc758\ub3c4 \ubd84\ub958\ub97c \ud1b5\ud574 \uc0ac\uc6a9\uc790\uc758 \uc9c8\ubb38\uc744 \uc815\ud655\ud558\uac8c \ud30c\uc545\ud558\uace0 \uc801\uc808\ud55c \uc751\ub2f5\uc744 \uc0dd\uc131\ud569\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, “\ud658\ubd88\ud558\uace0 \uc2f6\uc5b4\uc694”\ub77c\ub294 \uc9c8\ubb38\uc740 \ud658\ubd88 \uc808\ucc28\uc5d0 \ub300\ud55c \uc815\ubcf4\ub97c \uc81c\uacf5\ud574\uc57c \ud560 \ud544\uc694\uac00 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
6.2 \uc74c\uc131 \ube44\uc11c<\/h3>\n
\uc74c\uc131 \uc778\uc2dd \uc11c\ube44\uc2a4\uc5d0\uc11c\ub3c4 \uc758\ub3c4 \ubd84\ub958\uac00 \uc911\uc694\ud569\ub2c8\ub2e4. \uc0ac\uc6a9\uc790\uac00 \uc74c\uc131 \uba85\ub839\uc744 \ud1b5\ud574 \ud2b9\uc815 \uc791\uc5c5\uc744 \uc694\uccad\ud560 \ub54c, \uc774\ub7ec\ud55c \uc758\ub3c4\ub97c \uc774\ud574\ud558\uace0 \uc2e4\ud589\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, “\uc601\ud654 \uc608\uc57d\ud574 \uc918”\ub77c\ub294 \uc694\uccad\uc5d0 \ub300\ud574 \uc801\uc808\ud55c \uc751\ub2f5\uc744 \uc81c\uacf5\ud574\uc57c \ud569\ub2c8\ub2e4.<\/p>\n
6.3 \ub9ac\ubdf0 \ubd84\uc11d<\/h3>\n
\uc81c\ud488 \ub9ac\ubdf0\ub97c \ubd84\uc11d\ud558\uc5ec \uc0ac\uc6a9\uc790\uc758 \uae0d\uc815\uc801 \ub610\ub294 \ubd80\uc815\uc801 \uc758\ub3c4\ub97c \ud30c\uc545\ud558\uace0, \uc774\ub97c \ud1b5\ud574 \uc81c\ud488 \uac1c\uc120\uc774\ub098 \ub9c8\ucf00\ud305 \uc804\ub7b5\uc744 \uc218\ub9bd\ud558\ub294 \ub370 \ub3c4\uc6c0\uc744 \uc904 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
7. \uacb0\ub860<\/h2>\n
\ub525 \ub7ec\ub2dd\uc744 \uc774\uc6a9\ud55c \uc790\uc5f0\uc5b4 \ucc98\ub9ac, \ud2b9\ud788 \uc0ac\uc804 \ud6c8\ub828\ub41c \uc6cc\ub4dc \uc784\ubca0\ub529\uc744 \ud65c\uc6a9\ud55c \uc758\ub3c4 \ubd84\ub958\ub294 \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0\uc11c \ud601\uc2e0\uc801\uc778 \ubcc0\ud654\ub97c \uac00\uc838\uc654\uc2b5\ub2c8\ub2e4. \uc0ac\uc6a9\uc790\uc758 \uc758\ub3c4\ub97c \uc815\ud655\ud558\uac8c \uc774\ud574\ud568\uc73c\ub85c\uc368 \ucc57\ubd07, \uc74c\uc131 \ube44\uc11c \ub4f1\uc5d0\uc11c \uc0ac\uc6a9\uc790 \uacbd\ud5d8\uc744 \ud5a5\uc0c1\uc2dc\ud0a4\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc55e\uc73c\ub85c \ub354\uc6b1 \ubc1c\uc804\ud560 \uc790\uc5f0\uc5b4 \ucc98\ub9ac \uae30\uc220\ub4e4\uc774 \uae30\ub300\ub429\ub2c8\ub2e4.<\/p>\n
\ucc38\uace0 \ubb38\ud5cc<\/h2>\n\n- Mikolov, T., Sutskever, I., Chen, K., Corrado, G., & Dean, J. (2013). ‘Distributed Representations of Words and Phrases and their Compositionality’. In Advances in Neural Information Processing Systems, 26.<\/li>\n
- Pennington, J., Socher, R., & Manning, C. D. (2014). ‘Glove: Global Vectors for Word Representation’. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP).<\/li>\n
- Bojanowski, P., Grave, E., Mikolov, T., Sutskever, I., & Jozefowicz, R. (2017). ‘Enriching Word Vectors with Subword Information’. Transactions of the Association for Computational Linguistics, 5, 135-146.<\/li>\n<\/ul>\n
<\/body><\/p>\n","protected":false},"excerpt":{"rendered":"