\n \uc790\uc5f0\uc5b4 \ucc98\ub9ac\ub294 \uc778\uacf5\uc9c0\ub2a5(AI)\uc758 \ud55c \ubd84\uc57c\ub85c, \ucef4\ud4e8\ud130\uac00 \uc778\uac04\uc758 \uc5b8\uc5b4\ub97c \uc774\ud574\ud558\uace0 \ud574\uc11d\ud558\ub294 \uac83\uc744 \ubaa9\ud45c\ub85c \ud569\ub2c8\ub2e4. \ucd5c\uadfc \uba87 \ub144 \uac04 \ub525 \ub7ec\ub2dd \uae30\uc220\uc758 \ubc1c\uc804\uc73c\ub85c \uc778\ud574 \uc790\uc5f0\uc5b4 \ucc98\ub9ac(NLP)\ub294 \uadf8 \uc131\ub2a5\uc774 \ube44\uc57d\uc801\uc73c\ub85c \ud5a5\uc0c1\ub418\uc5c8\uc2b5\ub2c8\ub2e4. \ubcf8 \uae00\uc5d0\uc11c\ub294 BiLSTM-CRF \ubaa8\ub378\uc744 \ud65c\uc6a9\ud55c \uac1c\uccb4\uba85 \uc778\uc2dd(NER: Named Entity Recognition)\uc5d0 \ub300\ud574 \uc0c1\uc138\ud788 \uc124\uba85\ud558\uaca0\uc2b5\ub2c8\ub2e4.\n<\/p>\n
\n \uac1c\uccb4\uba85 \uc778\uc2dd\uc740 \uc8fc\uc5b4\uc9c4 \ud14d\uc2a4\ud2b8\uc5d0\uc11c \uace0\uc720\uba85\uc0ac, \uc7a5\uc18c\uba85, \ub0a0\uc9dc \ub4f1 \ud2b9\uc815 \uac1c\uccb4\ub97c \uc2dd\ubcc4\ud558\uace0 \ubd84\ub958\ud558\ub294 \uc791\uc5c5\uc785\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, “\uc774\uac15\uc778 \uc120\uc218\uac00 \ubc14\ub974\uc140\ub85c\ub098\uc5d0\uc11c \ud65c\uc57d\ud558\uace0 \uc788\ub2e4.”\ub77c\ub294 \ubb38\uc7a5\uc5d0\uc11c “\uc774\uac15\uc778”\uc740 \uc778\ubb3c, “\ubc14\ub974\uc140\ub85c\ub098”\ub294 \uc7a5\uc18c\ub85c \uc2dd\ubcc4\ub429\ub2c8\ub2e4. NER\uc740 \uc815\ubcf4 \ucd94\ucd9c, \uc9c8\uc758 \uc751\ub2f5 \uc2dc\uc2a4\ud15c, \ub300\ud654\ud615 AI \ub4f1 \ub2e4\uc591\ud55c NLP \uc560\ud50c\ub9ac\ucf00\uc774\uc158\uc5d0\uc11c \uc911\uc694\ud55c \uc5ed\ud560\uc744 \ud569\ub2c8\ub2e4.\n<\/p>\n
\n \uac1c\uccb4\uba85 \uc778\uc2dd\uc774 \uc911\uc694\ud55c \uc774\uc720\ub294 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4:<\/p>\n
\n BiLSTM-CRF\ub294 \uac1c\uccb4\uba85 \uc778\uc2dd \uc791\uc5c5\uc5d0\uc11c \ub110\ub9ac \uc0ac\uc6a9\ub418\ub294 \ubaa8\ub378\uc785\ub2c8\ub2e4. BiLSTM(Bidirectional Long Short-Term Memory)\uacfc CRF(Conditional Random Field)\uc758 \uacb0\ud569\uc740 \ubb38\ub9e5 \uc815\ubcf4\ub97c \ud6a8\uacfc\uc801\uc73c\ub85c \ud559\uc2b5\ud558\uace0, \uc608\uce21 \uacb0\uacfc\uc758 \uc77c\uad00\uc131\uc744 \ubcf4\uc7a5\ud569\ub2c8\ub2e4.\n<\/p>\n
\n LSTM(Long Short-Term Memory)\uc740 RNN(Recurrent Neural Network)\uc758 \ud55c \uc885\ub958\ub85c, \uae34 \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\ub97c \ucc98\ub9ac\ud558\ub294 \ub370 \uac15\ub825\ud55c \uc131\ub2a5\uc744 \ubcf4\uc778\ub2e4. LSTM\uc740 ‘\uc140 \uc0c1\ud0dc’\ub97c \uc720\uc9c0\ud558\uba70, \uc815\ubcf4\uc758 \ud750\ub984\uc744 \uc81c\uc5b4\ud558\ub294 \uac8c\uc774\ud2b8\ub97c \ud1b5\ud574 \uacfc\uac70\uc758 \uc815\ubcf4\ub97c \uae30\uc5b5\ud558\uac70\ub098 \uc78a\uc5b4\ubc84\ub9ac\ub294 \ubc29\uc2dd\uc73c\ub85c \uc791\ub3d9\ud569\ub2c8\ub2e4. \uc774\ub294 \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\uc5d0\uc11c \uc7a5\uae30 \uc758\uc874\uc131\uc744 \ud559\uc2b5\ud558\ub294 \ub370 \ub9e4\uc6b0 \ud6a8\uacfc\uc801\uc785\ub2c8\ub2e4.\n<\/p>\n
\n BiLSTM\uc740 \ub450 \uac1c\uc758 LSTM \ub808\uc774\uc5b4\ub97c \ud65c\uc6a9\ud558\uc5ec \uc2dc\ud000\uc2a4\ub97c \uc591 \ubc29\ud5a5\uc73c\ub85c \ucc98\ub9ac\ud569\ub2c8\ub2e4. \uc989, \ud55c \ubc29\ud5a5\uc740 \uc67c\ucabd\uc5d0\uc11c \uc624\ub978\ucabd\uc73c\ub85c, \ub2e4\ub978 \ubc29\ud5a5\uc740 \uc624\ub978\ucabd\uc5d0\uc11c \uc67c\ucabd\uc73c\ub85c \uc2dc\ud000\uc2a4\ub97c \uc77d\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ubc29\uc2dd\uc740 \uac01 \ub2e8\uc5b4\uac00 \uc8fc\ubcc0 \ub9e5\ub77d\uc744 \ubcf4\ub2e4 \uc798 \ubc18\uc601\ud560 \uc218 \uc788\ub3c4\ub85d \ud569\ub2c8\ub2e4.\n<\/p>\n
\n CRF\ub294 \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\uc5d0\uc11c\uc758 \uc758\uc874\uc131 \uad00\uacc4\ub97c \ubaa8\ub378\ub9c1\ud558\ub294 \ub370 \uc0ac\uc6a9\ub418\ub294 \uad6c\uc870\uc801 \uc608\uce21 \ubaa8\ub378\uc785\ub2c8\ub2e4. NER\uc640 \uac19\uc740 \ud0dc\uae45 \ubb38\uc81c\uc5d0\uc11c\ub294 \uac01 \ub2e8\uc5b4\uac00 \uc18d\ud55c \ud074\ub798\uc2a4\uc758 \uc870\uac74\ubd80 \ud655\ub960\uc744 \uace0\ub824\ud558\uc5ec \ucd5c\uc801\uc758 \ud0dc\uadf8 \uc2dc\ud000\uc2a4\ub97c \ucc3e\ub294 \ub370 \uc720\uc6a9\ud569\ub2c8\ub2e4. CRF\ub294 \uc608\uce21\uc758 \uc77c\uad00\uc131\uc744 \uc720\uc9c0\ud560 \uc218 \uc788\ub3c4\ub85d \ud558\uba70, \uc608\ub97c \ub4e4\uc5b4 “\ub274\uc695”\uc774\ub77c\ub294 \ub2e8\uc5b4\uac00 \ub3c4\uc2dc\uc784\uc744 \uc608\uce21\ud588\uc744 \ub54c, \uadf8 \ub2e4\uc74c \ub2e8\uc5b4\uac00 \uc7a5\uc18c\uc640 \uad00\ub828\ub418\uc5b4 \uc788\uc744 \ud655\ub960\uc744 \ub192\uc785\ub2c8\ub2e4.\n<\/p>\n
\n BiLSTM-CRF \ubaa8\ub378\uc740 \ub2e4\uc74c\uacfc \uac19\uc740 \uad6c\uc870\ub97c \uac00\uc9d1\ub2c8\ub2e4:<\/p>\n
\n \uc774\uc81c BiLSTM-CRF \ubaa8\ub378\uc744 \uad6c\ud604\ud558\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc774 \uad6c\ud604\uc5d0 \ud544\uc694\ud55c \uc8fc\uc694 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub294 \n NER \ubaa8\ub378\uc744 \ud559\uc2b5\ud558\uae30 \uc704\ud574\uc11c\ub294 \ub77c\ubca8\ub9c1\ub41c \ub370\uc774\ud130\uac00 \ud544\uc694\ud569\ub2c8\ub2e4. \ud754\ud788 \uc0ac\uc6a9\ub418\ub294 \ub370\uc774\ud130\uc14b\uc73c\ub85c\ub294 CoNLL-2003 \ub370\uc774\ud130\uc14b\uc774 \uc788\uc73c\uba70, \uc774 \ub370\uc774\ud130\ub294 \uac01 \ub2e8\uc5b4\uac00 \uc5b4\ub5a4 \uac1c\uccb4\uba85\uc5d0 \ud574\ub2f9\ud558\ub294\uc9c0\ub97c \ub2f4\uace0 \uc788\uc2b5\ub2c8\ub2e4. \ub370\uc774\ud130\ub294 \uc77c\ubc18\uc801\uc73c\ub85c \ud14d\uc2a4\ud2b8 \ud30c\uc77c\ub85c \uc81c\uacf5\ub418\uba70, \uac01 \uc904\uc5d0 \ub2e8\uc5b4\uc640 \uadf8 \ud0dc\uadf8\uac00 \uacf5\ubc31\uc73c\ub85c \uad6c\ubd84\ub418\uc5b4 \uc788\uc2b5\ub2c8\ub2e4.\n<\/p>\n \n \ub370\uc774\ud130 \uc804\ucc98\ub9ac\ub294 \ubb38\uc790\uc758 \uc815\uaddc\ud654, \ubd88\uc6a9\uc5b4 \uc81c\uac70, \ub2e8\uc5b4 \ubca1\ud130\ud654 \ub4f1 \uc5ec\ub7ec \ub2e8\uacc4\ub97c \ud3ec\ud568\ud569\ub2c8\ub2e4. \uc77c\ubc18\uc801\uc778 \uc804\ucc98\ub9ac \uacfc\uc815\uc740 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4:<\/p>\n \n \ubaa8\ub378\uc744 \ucef4\ud30c\uc77c\ud560 \ub54c \uc190\uc2e4 \ud568\uc218\ub85c\ub294 \uce74\ud14c\uace0\ub9ac \ud06c\ub85c\uc2a4 \uc5d4\ud2b8\ub85c\ud53c\ub97c \uc0ac\uc6a9\ud569\ub2c8\ub2e4. \ubaa8\ub378 \ud559\uc2b5\uc740 \ud6c8\ub828 \ub370\uc774\ud130 \uc138\ud2b8\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc218\ud589\ud569\ub2c8\ub2e4.\n<\/p>\n \n \ubaa8\ub378\uc758 \uc131\ub2a5\uc744 \ud3c9\uac00\ud558\uae30 \uc704\ud574 \ud63c\ub3d9 \ud589\ub82c, \uc815\ubc00\ub3c4, \uc7ac\ud604\uc728 \ub4f1\uc744 \ud655\uc778\ud569\ub2c8\ub2e4. \uc608\uce21 \ub610\ud55c \ub3d9\uc77c\ud55c \ubc29\uc2dd\uc73c\ub85c \uc9c4\ud589\ud558\uba70, \uc0c8\ub85c\uc6b4 \ubb38\uc7a5\uc744 \uc785\ub825\ubc1b\uc544 \uac1c\uccb4\uba85\uc744 \ucd94\ucd9c\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n<\/p>\n \n BiLSTM-CRF \ubaa8\ub378\uc740 \uac1c\uccb4\uba85 \uc778\uc2dd \uc791\uc5c5\uc744 \uc704\ud574 \ud6a8\uacfc\uc801\uc778 \uc811\uadfc \ubc29\uc2dd\uc744 \uc81c\uacf5\ud569\ub2c8\ub2e4. \ub525 \ub7ec\ub2dd \uae30\uc220\uacfc CRF\uc758 \uc870\ud654\ub97c \ud1b5\ud574, \uc6b0\ub9ac\ub294 \uc790\uc5f0\uc5b4\uc758 \ubcf5\uc7a1\uc131\uc744 \ub2e4\ub8f0 \uc218 \uc788\ub294 \uac15\ub825\ud55c \ub3c4\uad6c\ub97c \ud65c\uc6a9\ud560 \uc218 \uc788\uac8c \ub418\uc5c8\uc2b5\ub2c8\ub2e4. \ud5a5\ud6c4 \ub354 \ubc1c\uc804\ub41c \ubaa8\ub378\uc744 \ud1b5\ud574 \ub2e4\uc591\ud55c \uc5b8\uc5b4\uc640 \ub3c4\uba54\uc778\uc5d0\uc11c\ub3c4 \ub110\ub9ac \uc0ac\uc6a9\ub420 \uc218 \uc788\uae30\ub97c \uae30\ub300\ud569\ub2c8\ub2e4.\n<\/p>\n \n \ubcf8 \uae00\uc774 \ub525 \ub7ec\ub2dd \ubc0f NER\uc5d0 \ub300\ud55c \uc774\ud574\uc5d0 \ub3c4\uc6c0\uc774 \ub418\uc5c8\uae30\ub97c \ubc14\ub77c\uba70, \ucd94\uac00\uc801\uc778 \uc9c8\ubb38\uc774\ub098 \ub17c\uc758\uac00 \ud544\uc694\ud560 \uacbd\uc6b0 \ub313\uae00\ub85c \ub0a8\uaca8\uc8fc\uc2dc\uae30 \ubc14\ub78d\ub2c8\ub2e4.\n<\/p>\n <\/body><\/p>\n","protected":false},"excerpt":{"rendered":"TensorFlow<\/code>\uc640 Keras<\/code>\uc785\ub2c8\ub2e4.\n<\/p>\n3.1. \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac \uc124\uce58<\/h3>\n
\n
pip install tensorflow<\/code>\npip install keras<\/code>\n<\/pre>\n3.2. \ub370\uc774\ud130 \uc900\ube44<\/h3>\n
3.3. \ub370\uc774\ud130 \uc804\ucc98\ub9ac<\/h3>\n
\n
3.4. \ubaa8\ub378 \uad6c\uc131<\/h3>\n
\n
\nimport tensorflow as tf\nfrom tensorflow.keras.layers import Input, LSTM, Embedding, TimeDistributed, Dense, Bidirectional\nfrom tensorflow.keras.models import Model\n\ndef create_model(vocab_size, tag_size, embedding_dim=64, lstm_units=50):\n input = Input(shape=(None,))\n model = Embedding(input_dim=vocab_size, output_dim=embedding_dim, mask_zero=True)(input)\n model = Bidirectional(LSTM(units=lstm_units, return_sequences=True))(model)\n out = TimeDistributed(Dense(tag_size, activation=\"softmax\"))(model)\n return Model(input, out)\n<\/code>\n<\/pre>\n3.5. \ubaa8\ub378 \ucef4\ud30c\uc77c \ubc0f \ud559\uc2b5<\/h3>\n
\n
\nmodel.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])\nmodel.fit(X_train, y_train, batch_size=32, epochs=5, validation_data=(X_val, y_val))\n<\/code>\n<\/pre>\n4. \ubaa8\ub378 \ud3c9\uac00 \ubc0f \uc608\uce21<\/h2>\n
\n
\npredictions = model.predict(X_test)\n<\/code>\n<\/pre>\n5. \uacb0\ub860<\/h2>\n