\n \uc790\uc5f0\uc5b4 \ucc98\ub9ac(NLP, Natural Language Processing)\ub294 \uc778\uacf5\uc9c0\ub2a5(AI) \ubd84\uc57c\uc5d0\uc11c \uc911\uc694\ud558\uac8c \ub2e4\ub8e8\uc5b4\uc9c0\ub294 \ud55c \ubd84\uc57c\ub85c, \ucef4\ud4e8\ud130\uac00 \uc778\uac04\uc758 \uc5b8\uc5b4\ub97c \uc774\ud574\ud558\uace0 \ud574\uc11d\ud560 \uc218 \uc788\ub3c4\ub85d \ub3d5\uc2b5\ub2c8\ub2e4. \n \ud14d\uc2a4\ud2b8 \uc694\uc57d\uc740 \uc6d0\ubcf8 \ubb38\uc11c\uc758 \uc911\uc694\ud55c \uc815\ubcf4\ub97c \uac04\ub7b5\ud558\uac8c \uc694\uc57d\ud558\uc5ec \uc81c\uacf5\ud558\ub294 \uc791\uc5c5\uc744 \uc758\ubbf8\ud569\ub2c8\ub2e4. \n \ucd94\ucd9c\uc801 \uc694\uc57d\uc740 \ubb38\uc11c\uc758 \ub0b4\uc6a9\uc744 \ubd84\uc11d\ud558\uc5ec \uac00\uc7a5 \uc911\uc694\ud55c \ubb38\uc7a5\ub4e4\uc744 \uc120\ud0dd\ud558\ub294 \ubc29\uc2dd\uc785\ub2c8\ub2e4. \uc774 \uae30\ubc95\uc740 \ubcf4\ud1b5 \ub2e4\uc74c\uacfc \uac19\uc740 \ubc29\ubc95\uc744 \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n<\/p>\n \n \ucd94\uc0c1\uc801 \uc694\uc57d\uc740 \uc6d0\ubb38\uc744 \uae30\ubc18\uc73c\ub85c \uc0c8\ub85c\uc6b4 \ub0b4\uc6a9\uc744 \uc0dd\uc131\ud558\ub294 \uacfc\uc815\uc744 \uc758\ubbf8\ud569\ub2c8\ub2e4. \uc774\ub294 \ub354 \ucc3d\uc758\uc801\uc774\uace0 \ub17c\ub9ac\uc801\uc778 \uc694\uc57d\uc744 \uac00\ub2a5\ud558\uac8c \ud569\ub2c8\ub2e4. \n \ub525 \ub7ec\ub2dd\uc740 \uc778\uacf5\uc2e0\uacbd\ub9dd\uc744 \uae30\ubc18\uc73c\ub85c \ud558\ub294 \uba38\uc2e0\ub7ec\ub2dd \uae30\ubc95\uc73c\ub85c, \ub300\ub7c9\uc758 \ub370\uc774\ud130\ub97c \ud1b5\ud574 \ud328\ud134\uc744 \ud559\uc2b5\ud558\ub294 \ub370 \ucd5c\uc801\ud654\ub418\uc5b4 \uc788\uc2b5\ub2c8\ub2e4. \n \uc804\ud1b5\uc801\uc778 \uc778\uacf5\uc2e0\uacbd\ub9dd\uc740 \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\ub97c \ucc98\ub9ac\ud558\ub294 \ub370 \ud55c\uacc4\uac00 \uc788\uc9c0\ub9cc, \uc21c\ud658 \uc2e0\uacbd\ub9dd(RNN, Recurrent Neural Network)\uc740 \uacfc\uac70 \uc815\ubcf4\ub97c \uae30\uc5b5\ud560 \uc218 \uc788\ub3c4\ub85d \uc124\uacc4\ub418\uc5c8\uc2b5\ub2c8\ub2e4. \n \ucd5c\uadfc NLP\uc5d0\uc11c \ud601\uc2e0\uc801\uc778 \ubc1c\uc804\uc744 \uc774\ub8ec \uac83\uc740 \ubc14\ub85c Transformer \ubaa8\ub378<\/strong>\uc785\ub2c8\ub2e4. \uc774 \ubaa8\ub378\uc740 RNN\uc774\ub098 LSTM\uacfc \ub2ec\ub9ac \uc804\uccb4 \ubb38\uc7a5\uc744 \ud55c \ubc88\uc5d0 \ucc98\ub9ac\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \n \uc5b4\ud150\uc158(attention) \uba54\ucee4\ub2c8\uc998\uc740 \uc785\ub825\uc758 \uac01 \ubd80\ubd84\uc5d0 \ucc28\ubcc4\uc801\uc778 \uac00\uc911\uce58\ub97c \ubd80\uc5ec\ud574 \uc815\ubcf4\ub97c \uc120\ud0dd\uc801\uc73c\ub85c \uac15\uc870\ud558\ub294 \ubc29\uc2dd\uc785\ub2c8\ub2e4. \n \uc5b4\ud150\uc158 \uba54\ucee4\ub2c8\uc998\uc740 \ub2e4\uc74c \uc138 \uac00\uc9c0 \uc8fc\uc694 \uad6c\uc131 \uc694\uc18c\ub85c \uc774\ub8e8\uc5b4\uc838 \uc788\uc2b5\ub2c8\ub2e4.\n<\/p>\n \n \uc774 \uc138 \uac00\uc9c0 \uc694\uc18c\ub97c \ubc14\ud0d5\uc73c\ub85c, \uc77c\uc885\uc758 \uac00\uc911\ud569\uc744 \ud1b5\ud574 \ucd5c\uc885 \ucd9c\ub825\uc744 \uc0dd\uc131\ud569\ub2c8\ub2e4.\n<\/p>\n \n \ud14d\uc2a4\ud2b8 \uc694\uc57d\uc744 \uc704\ud55c \ub525 \ub7ec\ub2dd \ubaa8\ub378\uc740 \uc8fc\ub85c seq2seq \uc544\ud0a4\ud14d\ucc98\ub97c \uc0ac\uc6a9\ud569\ub2c8\ub2e4. \n \ud14d\uc2a4\ud2b8 \uc694\uc57d\uc744 \uc704\ud574 \uc900\ube44\ud560 \ub370\uc774\ud130\ub294 \uc77c\ubc18\uc801\uc73c\ub85c \uc6d0\ubcf8 \ubb38\uc7a5\uacfc \ud574\ub2f9 \ubb38\uc7a5\uc744 \uc694\uc57d\ud55c \ub370\uc774\ud130 \uc30d\uc785\ub2c8\ub2e4. \n \uae30\ubcf8\uc801\uc73c\ub85c seq2seq \uad6c\uc870\ub294 \uc778\ucf54\ub354\uc640 \ub514\ucf54\ub354\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4. \uc778\ucf54\ub354\ub294 \uc785\ub825 \ubb38\uc7a5\uc744 \ubc1b\uc544 \uace0\ucc28\uc6d0 \ubca1\ud130\ub85c \ubcc0\ud658\ud558\uace0, \ub514\ucf54\ub354\ub294 \uc774\ub97c \uae30\ubc18\uc73c\ub85c \uc694\uc57d\uc744 \uc0dd\uc131\ud569\ub2c8\ub2e4.\n<\/p>\n \n \ubaa8\ub378 \ud6c8\ub828\uc744 \uc704\ud574 \uc190\uc2e4 \ud568\uc218\ub97c \uc815\uc758\ud558\uace0, \uc635\ud2f0\ub9c8\uc774\uc800\ub97c \uc124\uc815\ud569\ub2c8\ub2e4. \n \ubaa8\ub378\uc758 \uc131\ub2a5\uc744 \ud3c9\uac00\ud558\uae30 \uc704\ud574 BLEU( Bilingual Evaluation Understudy) \uc810\uc218\ub97c \uc790\uc8fc \uc0ac\uc6a9\ud569\ub2c8\ub2e4. \n \ub525 \ub7ec\ub2dd\uacfc \uc5b4\ud150\uc158 \uba54\ucee4\ub2c8\uc998\uc744 \ud65c\uc6a9\ud55c \ud14d\uc2a4\ud2b8 \uc694\uc57d \uae30\uc220\uc740 \uc774\ub860\uc801\uc73c\ub85c\ub098 \uc2e4\uc6a9\uc801\uc73c\ub85c \ub9ce\uc740 \uac00\ub2a5\uc131\uc744 \uac00\uc9c0\uace0 \uc788\uc2b5\ub2c8\ub2e4. <\/body><\/p>\n","protected":false},"excerpt":{"rendered":"
\n \ucd5c\uadfc \uba87 \ub144 \ub3d9\uc548, \ub525 \ub7ec\ub2dd\uc758 \ubc1c\uc804\uc740 NLP\uc758 \ub9ce\uc740 \uacfc\uc81c\ub97c \ud68d\uae30\uc801\uc73c\ub85c \ud574\uacb0\ud558\ub294 \ub370 \ud06c\uac8c \uae30\uc5ec\ud588\uc2b5\ub2c8\ub2e4.
\n \uadf8 \uc911 \ud558\ub098\uac00 \ubc14\ub85c \ud14d\uc2a4\ud2b8 \uc694\uc57d(Text Summarization)<\/strong>\uc785\ub2c8\ub2e4. \ubcf8 \uae00\uc5d0\uc11c\ub294 \ub525 \ub7ec\ub2dd\uc744 \ud65c\uc6a9\ud55c \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc758 \uae30\ubcf8 \uac1c\ub150\uacfc \ud568\uaed8, \uc5b4\ud150\uc158 \uba54\ucee4\ub2c8\uc998\uc744 \uc774\uc6a9\ud55c \ud14d\uc2a4\ud2b8 \uc694\uc57d\uc758 \uc6d0\ub9ac\uc640 \uad6c\ud604 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc790\uc138\ud788 \uc124\uba85\ud558\uaca0\uc2b5\ub2c8\ub2e4.\n<\/p>\n1. \ud14d\uc2a4\ud2b8 \uc694\uc57d\uc758 \uc774\ud574<\/h2>\n
\n \uc774\ub294 \uc815\ubcf4 \uacfc\ub2e4\ub85c \uc778\ud55c \ubb38\uc81c\ub97c \ud574\uacb0\ud558\uace0, \ub3c5\uc790\uac00 \uc2e0\uc18d\ud558\uac8c \uc911\uc694\ud55c \ub0b4\uc6a9\uc744 \ud30c\uc545\ud560 \uc218 \uc788\ub3c4\ub85d \ub3d5\uc2b5\ub2c8\ub2e4.\n<\/p>\n\n
1.1 \ucd94\ucd9c\uc801 \uc694\uc57d<\/h3>\n
\n
1.2 \ucd94\uc0c1\uc801 \uc694\uc57d<\/h3>\n
\n \ub525 \ub7ec\ub2dd \ubaa8\ub378, \ud2b9\ud788 \uc2dc\ud000\uc2a4-\ud22c-\uc2dc\ud000\uc2a4(seq2seq) \uc544\ud0a4\ud14d\ucc98\uc640 \uc5b4\ud150\uc158 \uba54\ucee4\ub2c8\uc998\uc774 \uc774 \uacfc\uc815\uc5d0\uc11c \uc911\uc694\ud55c \uc5ed\ud560\uc744 \ud569\ub2c8\ub2e4.\n<\/p>\n2. \ub525 \ub7ec\ub2dd\uacfc NLP<\/h2>\n
\n \uc790\uc5f0\uc5b4 \ucc98\ub9ac\uc5d0\uc11c \uc774\ub7ec\ud55c \ub525 \ub7ec\ub2dd \uae30\ubc95\uc774 \uc0ac\uc6a9\ub418\uba74\uc11c \uc815\ubcf4\uc758 \uad6c\uc870\ub97c \uc774\ud574\ud558\uace0 \ubb38\uc7a5\uc744 \ucc98\ub9ac\ud558\ub294 \ub370 \uc788\uc5b4 \ud070 \ud601\uc2e0\uc774 \uc788\uc5c8\uc2b5\ub2c8\ub2e4.\n<\/p>\n2.1 RNN\uacfc LSTM<\/h3>\n
\n \uadf8\ub7ec\ub098 RNN\uc740 \uae34 \uc2dc\ud000\uc2a4\uc5d0 \ub300\ud55c \ud559\uc2b5\uc5d0 \uc5b4\ub824\uc6c0\uc744 \uacaa\ub294 \ubb38\uc81c\uc810\uc774 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub97c \ud574\uacb0\ud558\uae30 \uc704\ud574 \uac1c\ubc1c\ub41c \uac83\uc774 \ubc14\ub85c LSTM (Long Short-Term Memory)<\/strong>\uc785\ub2c8\ub2e4.\n<\/p>\n\n
2.2 Transformer \ubaa8\ub378<\/h3>\n
\n Transformer\uc758 \ud575\uc2ec \uad6c\uc131 \uc694\uc18c\ub294 \uc5b4\ud150\uc158 \uba54\ucee4\ub2c8\uc998\uc785\ub2c8\ub2e4.\n<\/p>\n3. \uc5b4\ud150\uc158 \uba54\ucee4\ub2c8\uc998<\/h2>\n
\n \uc774 \ubc29\ubc95\uc740 \uae34 \ubb38\uc7a5\uc758 \uc815\ubcf4\uac00 \uc11c\ub85c \ub2e4\ub974\uac8c \uc911\uc694\ud560 \uc218 \uc788\ub2e4\ub294 \uc810\uc744 \uac10\uc548\ud558\uc5ec, \ubcf4\ub2e4 \ud6a8\uc728\uc801\uc73c\ub85c \uc815\ubcf4\ub97c \ucc98\ub9ac\ud558\ub294 \ub370 \ub3c4\uc6c0\uc744 \uc90d\ub2c8\ub2e4.\n<\/p>\n3.1 \uc5b4\ud150\uc158\uc758 \uc6d0\ub9ac<\/h3>\n
\n
3.2 \uc5b4\ud150\uc158\uc758 \uc885\ub958<\/h3>\n
\n
4. \ud14d\uc2a4\ud2b8 \uc694\uc57d\uc744 \uc704\ud55c \ubaa8\ub378 \uad6c\ud604<\/h2>\n
\n \uc774 \ubaa8\ub378\uc740 \uc785\ub825 \uc2dc\ud000\uc2a4\uc640 \ucd9c\ub825 \uc2dc\ud000\uc2a4 \uac04\uc758 \uad00\uacc4\ub97c \ud559\uc2b5\ud560 \uc218 \uc788\ub3c4\ub85d \ud569\ub2c8\ub2e4.\n<\/p>\n4.1 \ub370\uc774\ud130 \uc900\ube44<\/h3>\n
\n \ub300\ub7c9\uc758 \ub370\uc774\ud130\uc14b\uc774 \ud544\uc694\ud558\uba70, \ub274\uc2a4 \uae30\uc0ac, \ub17c\ubb38 \ub4f1\uc758 \ub2e4\uc591\ud55c \uc18c\uc2a4\ub97c \ud65c\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n<\/p>\n4.2 \ubaa8\ub378 \uc544\ud0a4\ud14d\ucc98<\/h3>\n
\nclass Seq2SeqModel(nn.Module):\n def __init__(self, encoder, decoder):\n super(Seq2SeqModel, self).__init__()\n self.encoder = encoder\n self.decoder = decoder\n\n def forward(self, src, trg):\n encoder_output = self.encoder(src)\n decoder_output = self.decoder(trg, encoder_output)\n return decoder_output\n<\/code><\/pre>\n4.3 \ud6c8\ub828 \uacfc\uc815<\/h3>\n
\n \uc77c\ubc18\uc801\uc73c\ub85c \uc0ac\uc6a9\ub418\ub294 \uc190\uc2e4 \ud568\uc218\ub294 \uad50\ucc28 \uc5d4\ud2b8\ub85c\ud53c \uc190\uc2e4(cross-entropy loss)\uc774\uba70, Adam \uc635\ud2f0\ub9c8\uc774\uc800\uac00 \uc790\uc8fc \uc0ac\uc6a9\ub429\ub2c8\ub2e4.\n<\/p>\n\ncriterion = nn.CrossEntropyLoss()\noptimizer = torch.optim.Adam(model.parameters(), lr=0.001)\n\nfor epoch in range(num_epochs):\n model.train()\n optimizer.zero_grad()\n outputs = model(src, trg)\n loss = criterion(outputs, trg)\n loss.backward()\n optimizer.step()\n<\/code><\/pre>\n5. \uc131\ub2a5 \ud3c9\uac00<\/h2>\n
\n BLEU \uc810\uc218\ub294 \ubaa8\ub378\uc774 \uc0dd\uc131\ud55c \uc694\uc57d\uacfc \uc2e4\uc81c \uc694\uc57d \uac04\uc758 \uc720\uc0ac\uc131\uc744 \uce21\uc815\ud558\ub294 \uc9c0\ud45c\ub85c, 0\uc5d0\uc11c 1 \uc0ac\uc774\uc758 \uac12\uc744 \uac00\uc9d1\ub2c8\ub2e4.
\n 1\uc5d0 \uac00\uae4c\uc6b8\uc218\ub85d \uc88b\uc740 \uc131\ub2a5\uc73c\ub85c \uac04\uc8fc\ub429\ub2c8\ub2e4.\n<\/p>\n5.1 BLEU \uc810\uc218 \uacc4\uc0b0<\/h3>\n
\nfrom nltk.translate.bleu_score import sentence_bleu\n\nreference = [actual_summary.split()]\ncandidate = produced_summary.split()\n\nbleu_score = sentence_bleu(reference, candidate)\n<\/code><\/pre>\n6. \uacb0\ub860<\/h2>\n
\n \uc55e\uc73c\ub85c\uc758 \uc5f0\uad6c\uc640 \ubc1c\uc804\uc744 \ud1b5\ud574 \uc774 \uae30\uc220\uc774 \ub354\uc6b1 \ud655\uc0b0\ub418\uace0, \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0\uc11c \ud65c\uc6a9\ub420 \uc218 \uc788\uae30\ub97c \uae30\ub300\ud569\ub2c8\ub2e4.
\n \uc774 \uae00\uc5d0\uc11c\ub294 \uae30\ucd08\uc801\uc778 \uac1c\ub150\ubd80\ud130 \ubaa8\ub378 \uad6c\ud604\uae4c\uc9c0\uc758 \uacfc\uc815\uc744 \uc124\uba85\ud558\uc600\uc73c\uba70, \ub3c5\uc790 \uc5ec\ub7ec\ubd84\uc774 \uc774\ub97c \ubc14\ud0d5\uc73c\ub85c \uc2e4\uc81c \ud504\ub85c\uc81d\ud2b8\uc5d0 \uc801\uc6a9\ud558\uae30\ub97c \ubc14\ub78d\ub2c8\ub2e4.\n<\/p>\n