{"id":30154,"date":"2024-10-28T03:19:50","date_gmt":"2024-10-28T03:19:50","guid":{"rendered":"http:\/\/atmokpo.com\/w\/?p=30154"},"modified":"2024-11-26T06:49:41","modified_gmt":"2024-11-26T06:49:41","slug":"%eb%94%a5%eb%9f%ac%eb%8b%9d-%ed%8c%8c%ec%9d%b4%ed%86%a0%ec%b9%98-%ea%b0%95%ec%a2%8c-%ed%9a%9f%ec%88%98-%ea%b8%b0%eb%b0%98-%ec%9e%84%eb%b2%a0%eb%94%a9","status":"publish","type":"post","link":"https:\/\/atmokpo.com\/w\/30154\/","title":{"rendered":"\ub525\ub7ec\ub2dd \ud30c\uc774\ud1a0\uce58 \uac15\uc88c, \ud69f\uc218 \uae30\ubc18 \uc784\ubca0\ub529"},"content":{"rendered":"

<\/p>\n

\ub525\ub7ec\ub2dd \ubd84\uc57c\uc5d0\uc11c \uc784\ubca0\ub529(embedding)\uc740 \ub370\uc774\ud130\uc758 \ud488\uc9c8\uc744 \ub192\uc774\uace0, \ub354 \ub098\uc740 \ud559\uc2b5 \uacb0\uacfc\ub97c \uc5bb\uae30 \uc704\ud574 \ub9e4\uc6b0 \uc720\uc6a9\ud55c \uae30\ubc95\uc785\ub2c8\ub2e4. \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 \ud69f\uc218 \uae30\ubc18 \uc784\ubca0\ub529\uc5d0 \ub300\ud574 \uc18c\uac1c\ud558\uace0, PyTorch\ub97c \uc0ac\uc6a9\ud574 \uc774\ub97c \uc5b4\ub5bb\uac8c \uad6c\ud604\ud558\ub294\uc9c0\uc5d0 \ub300\ud574 \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n

1. \uc784\ubca0\ub529\uc774\ub780 \ubb34\uc5c7\uc778\uac00?<\/h2>\n

\uc784\ubca0\ub529\uc740 \uace0\ucc28\uc6d0 \ub370\uc774\ud130\ub97c \uc800\ucc28\uc6d0\uc73c\ub85c \ubcc0\ud658\ud558\uc5ec \uc758\ubbf8\ub97c \uac00\uc9c4 \ubca1\ud130 \uacf5\uac04\uc744 \ub9cc\ub4dc\ub294 \ubc29\ubc95\uc785\ub2c8\ub2e4. \ud2b9\ud788 \uc790\uc5f0\uc5b4 \ucc98\ub9ac, \ucd94\ucc9c \uc2dc\uc2a4\ud15c, \uc774\ubbf8\uc9c0 \ucc98\ub9ac \ubd84\uc57c\uc5d0\uc11c \ub110\ub9ac \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, \ub2e8\uc5b4\ub97c \ubca1\ud130\ub85c \ud45c\ud604\ud558\uba74 \ub2e8\uc5b4\ub4e4 \uac04\uc758 \uc758\ubbf8\uc801 \uc720\uc0ac\uc131\uc744 \uacc4\uc0b0\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n

2. \ud69f\uc218 \uae30\ubc18 \uc784\ubca0\ub529\uc758 \uac1c\ub150<\/h2>\n

\ud69f\uc218 \uae30\ubc18 \uc784\ubca0\ub529\uc740 \uc8fc\uc5b4\uc9c4 \ub370\uc774\ud130\uc758 \ucd9c\ud604 \ube48\ub3c4(count)\ub97c \uae30\ubc18\uc73c\ub85c \ub2e8\uc5b4 \ub610\ub294 \uac1d\uccb4\ub97c \uc784\ubca0\ub529\ud558\ub294 \ubc29\ubc95\uc785\ub2c8\ub2e4. \uc774 \ubc29\ubc95\uc740 \uc8fc\ub85c \ubb38\uc11c\uc5d0\uc11c \ub2e8\uc5b4\uc758 \ucd9c\ud604 \ube48\ub3c4\ub97c \uae30\ubc18\uc73c\ub85c \ub2e8\uc5b4 \uac04\uc758 \uad00\uacc4\ub97c \ud1a0\ub300\ub85c \uc784\ubca0\ub529\uc744 \uc0dd\uc131\ud569\ub2c8\ub2e4. \uac00\uc7a5 \ub110\ub9ac \uc54c\ub824\uc9c4 \ubc29\ubc95\uc740 TF-IDF(\uc6a9\uc5b4 \ube48\ub3c4-\uc5ed \ubb38\uc11c \ube48\ub3c4)\uc640 \uac19\uc740 \uc811\uadfc \ubc29\uc2dd\uc785\ub2c8\ub2e4.<\/p>\n

2.1. TF-IDF\uc758 \uae30\ubcf8 \uac1c\ub150<\/h3>\n

TF-IDF\ub294 \ubb38\uc11c \ub0b4\uc758 \ud2b9\uc815 \ub2e8\uc5b4\uc758 \uc911\uc694\ub3c4\ub97c \ud3c9\uac00\ud558\ub294 \ubc29\ubc95\uc73c\ub85c, \ub2e8\uc21c\ud788 \ub2e8\uc5b4\uc758 \ube48\ub3c4(frequency)\ub97c \ube44\uad50\ud558\ub294 \uac83\ubcf4\ub2e4 \ub354 \uc720\uc6a9\ud55c \uc815\ubcf4\ub97c \uc81c\uacf5\ud569\ub2c8\ub2e4. TF\ub294 ‘\uc6a9\uc5b4 \ube48\ub3c4(Term Frequency)’\ub97c \uc758\ubbf8\ud558\uace0, IDF\ub294 ‘\uc5ed \ubb38\uc11c \ube48\ub3c4(Inverse Document Frequency)’\ub97c \uc758\ubbf8\ud569\ub2c8\ub2e4.<\/p>\n

2.2. TF-IDF \uacc4\uc0b0<\/h3>\n

TF-IDF\ub294 \ub2e4\uc74c\uacfc \uac19\uc774 \uacc4\uc0b0\ub429\ub2c8\ub2e4:<\/p>\n

\nTF = (\ud574\ub2f9 \ub2e8\uc5b4\uc758 \ubb38\uc11c \ub0b4 \ucd9c\ud604 \ud69f\uc218) \/ (\ubb38\uc11c \ub0b4 \uc804\uccb4 \ub2e8\uc5b4 \uc218)\nIDF = log(\ucd1d \ubb38\uc11c \uc218 \/ (\ud574\ub2f9 \ub2e8\uc5b4\uac00 \ucd9c\ud604\ud55c \ubb38\uc11c \uc218 + 1))\nTF-IDF = TF * IDF\n<\/code><\/pre>\n
3. PyTorch\ub85c \ud69f\uc218 \uae30\ubc18 \uc784\ubca0\ub529 \uad6c\ud604\ud558\uae30<\/h2>\n
\uc774\uc81c PyTorch\ub97c \uc0ac\uc6a9\ud558\uc5ec \ud69f\uc218 \uae30\ubc18 \uc784\ubca0\ub529\uc744 \uad6c\ud604\ud558\ub294 \ubc29\ubc95\uc744 \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc608\uc81c\ub85c\ub294 \uac04\ub2e8\ud55c \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\uc14b\uc744 \uc0ac\uc6a9\ud558\uc5ec TF-IDF \uc784\ubca0\ub529\uc744 \uacc4\uc0b0\ud574 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
3.1. \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac \uc124\uce58<\/h3>\n
\npip install torch scikit-learn numpy pandas\n<\/code><\/pre>\n
3.2. \ub370\uc774\ud130 \uc900\ube44<\/h3>\n
\uc6b0\uc120 \uc0ac\uc6a9\ud560 \uac04\ub2e8\ud55c \uc608\uc81c \ub370\uc774\ud130\ub97c \uc0dd\uc131\ud558\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
\nimport pandas as pd\n\n# \uc608\uc81c \ub370\uc774\ud130 \uc0dd\uc131\ndata = {\n    'text': [\n        '\uc0ac\uacfc\ub294 \ub9db\uc788\ub2e4',\n        '\ubc14\ub098\ub098\ub294 \ub178\ub780\uc0c9\uc774\ub2e4',\n        '\uc0ac\uacfc\uc640 \ubc14\ub098\ub098\ub294 \uacfc\uc77c\uc774\ub2e4',\n        '\uc0ac\uacfc\ub294 \ube44\ud0c0\ubbfc\uc774 \ub9ce\ub2e4',\n        '\ubc14\ub098\ub098 \uacfc\uc77c\uc740 \uc5d0\ub108\uc9c0\uc6d0\uc774\ub2e4'\n    ]\n}\n\ndf = pd.DataFrame(data)\nprint(df)\n<\/code><\/pre>\n
3.3. TF-IDF \ubca1\ud130\ud654<\/h3>\n
\uc774\uc81c \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\ub97c TF-IDF \ubca1\ud130\ub85c \ubcc0\ud658\ud569\ub2c8\ub2e4. \uc774\ub97c \uc704\ud574 sklearn<\/code>\uc758 TfidfVectorizer<\/code>\ub97c \uc0ac\uc6a9\ud560 \uac83\uc785\ub2c8\ub2e4.<\/p>\n
\nfrom sklearn.feature_extraction.text import TfidfVectorizer\n\n# TF-IDF \ubca1\ud130 \uc0dd\uc131\nvectorizer = TfidfVectorizer()\ntfidf_matrix = vectorizer.fit_transform(df['text'])\n\n# \uacb0\uacfc \ucd9c\ub825\ntfidf_df = pd.DataFrame(tfidf_matrix.toarray(), columns=vectorizer.get_feature_names_out())\nprint(tfidf_df)\n<\/code><\/pre>\n
3.4. PyTorch \ub370\uc774\ud130\uc14b \ubc0f DataLoader \uc900\ube44<\/h3>\n
PyTorch\uc5d0\uc11c \ub370\uc774\ud130\ub97c \ucc98\ub9ac\ud560 \uc218 \uc788\ub3c4\ub85d Dataset<\/code> \ubc0f DataLoader<\/code>\ub97c \uc815\uc758\ud558\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
\nimport torch\nfrom torch.utils.data import Dataset, DataLoader\n\nclass TFIDFDataset(Dataset):\n    def __init__(self, tfidf_matrix):\n        self.tfidf_matrix = tfidf_matrix\n\n    def __len__(self):\n        return self.tfidf_matrix.shape[0]\n\n    def __getitem__(self, idx):\n        return torch.tensor(self.tfidf_matrix[idx], dtype=torch.float32)\n\n# \ub370\uc774\ud130\uc14b \uc0dd\uc131\ntfidf_dataset = TFIDFDataset(tfidf_df.values)\ndata_loader = DataLoader(tfidf_dataset, batch_size=2, shuffle=True)\n<\/code><\/pre>\n
3.5. \ubaa8\ub378 \uc815\uc758<\/h3>\n
\uc774\uc81c \uac04\ub2e8\ud55c \uc2e0\uacbd\ub9dd \ubaa8\ub378\uc744 \uc815\uc758\ud558\uc5ec \ud69f\uc218 \uae30\ubc18 \uc784\ubca0\ub529\uc744 \ud559\uc2b5\ud574 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
\nimport torch.nn as nn\n\nclass SimpleNN(nn.Module):\n    def __init__(self, input_dim, hidden_dim, output_dim):\n        super(SimpleNN, self).__init__()\n        self.fc1 = nn.Linear(input_dim, hidden_dim)\n        self.fc2 = nn.Linear(hidden_dim, output_dim)\n\n    def forward(self, x):\n        x = torch.relu(self.fc1(x))\n        x = self.fc2(x)\n        return x\n\n# \ubaa8\ub378 \ucd08\uae30\ud654\ninput_dim = tfidf_df.shape[1]\nhidden_dim = 4\noutput_dim = 2  # \uc608\ub97c \ub4e4\uc5b4, \ub450 \uac1c\uc758 \ud074\ub798\uc2a4\ub85c \ubd84\ub958\nmodel = SimpleNN(input_dim, hidden_dim, output_dim)\n<\/code><\/pre>\n
3.6. \ud559\uc2b5 \uacfc\uc815 \uc124\uc815<\/h3>\n
\ubaa8\ub378\uc744 \ud559\uc2b5\ud558\uae30 \uc704\ud574 \uc190\uc2e4 \ud568\uc218\uc640 \ucd5c\uc801\ud654 \uc54c\uace0\ub9ac\uc998\uc744 \uc815\uc758\ud569\ub2c8\ub2e4.<\/p>\n
\ncriterion = nn.CrossEntropyLoss()\noptimizer = torch.optim.Adam(model.parameters(), lr=0.001)\n\n# \ud559\uc2b5 \uacfc\uc815\nnum_epochs = 100\nfor epoch in range(num_epochs):\n    for batch in data_loader:\n        optimizer.zero_grad()\n        outputs = model(batch)\n        labels = torch.tensor([0, 1])  # \ub354\ubbf8 \ub808\uc774\ube14\n        loss = criterion(outputs, labels)\n        loss.backward()\n        optimizer.step()\n    \n    if (epoch+1) % 10 == 0:\n        print(f'Epoch [{epoch+1}\/{num_epochs}], Loss: {loss.item():.4f}')\n<\/code><\/pre>\n
4. \uacb0\ub860<\/h2>\n
\ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 \ud69f\uc218 \uae30\ubc18 \uc784\ubca0\ub529\uc758 \uac1c\ub150\uacfc \uc774\ub97c PyTorch\ub85c \uad6c\ud604\ud558\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc0b4\ud3b4\ubcf4\uc558\uc2b5\ub2c8\ub2e4. TF-IDF\ub97c \uc0ac\uc6a9\ud558\uc5ec \ub2e8\uc21c\ud55c \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\uc14b\uc758 \uc784\ubca0\ub529\uc744 \uc0dd\uc131\ud558\uace0, \uac04\ub2e8\ud55c \uc2e0\uacbd\ub9dd \ubaa8\ub378\uc744 \uc815\uc758\ud558\uc5ec \ud559\uc2b5\ud558\ub294 \uacfc\uc815\uc744 \ubcf4\uc5ec\ub4dc\ub838\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \uc784\ubca0\ub529 \uae30\ubc95\uc740 \uc790\uc5f0\uc5b4 \ucc98\ub9ac \ubc0f \ub370\uc774\ud130 \ubd84\uc11d\uc5d0\uc11c \ub9e4\uc6b0 \uc720\uc6a9\ud558\uac8c \uc0ac\uc6a9\ub420 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\ucc38\uace0 \ubb38\ud5cc<\/h3>\n