{"id":30156,"date":"2024-10-28T03:19:51","date_gmt":"2024-10-28T03:19:51","guid":{"rendered":"http:\/\/atmokpo.com\/w\/?p=30156"},"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-%ec%98%88%ec%b8%a1-%ea%b8%b0%eb%b0%98-%ec%9e%84%eb%b2%a0%eb%94%a9","status":"publish","type":"post","link":"https:\/\/atmokpo.com\/w\/30156\/","title":{"rendered":"\ub525\ub7ec\ub2dd \ud30c\uc774\ud1a0\uce58 \uac15\uc88c, \ud69f\uc218 \uc608\uce21 \uae30\ubc18 \uc784\ubca0\ub529"},"content":{"rendered":"

<\/p>\n

\n

\uc774 \uae00\uc5d0\uc11c\ub294 \ub525\ub7ec\ub2dd\uc758 \ud55c \ubd84\uc57c\uc778 \uc784\ubca0\ub529(Embedding)\uc5d0 \uad00\ud558\uc5ec \uc54c\uc544\ubcf4\uace0, \ud2b9\ud788 \ud69f\uc218 \uae30\ubc18\uacfc \uc608\uce21 \uae30\ubc18 \uc784\ubca0\ub529 \uae30\ubc95\uc5d0 \ub300\ud574 \uc790\uc138\ud788 \uc124\uba85\ud569\ub2c8\ub2e4. \ub610\ud55c, \ud30c\uc774\ud1a0\uce58(PyTorch) \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ud65c\uc6a9\ud558\uc5ec \uc774\ub97c \uad6c\ud604\ud558\ub294 \uc608\uc81c \ucf54\ub4dc\ub97c \uc81c\uacf5\ud558\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n<\/header>\n

\n

1. \uc784\ubca0\ub529\uc774\ub780?<\/h2>\n

\n \uc784\ubca0\ub529\uc740 \uace0\ucc28\uc6d0 \ub370\uc774\ud130\ub97c \uc800\ucc28\uc6d0\uc73c\ub85c \ubcc0\ud658\ud558\uc5ec \uc758\ubbf8\ub97c \ubcf4\uc874\ud558\ub294 \ubc29\ubc95\uc744 \uc758\ubbf8\ud569\ub2c8\ub2e4. \ud2b9\ud788 \uc790\uc5f0\uc5b4 \ucc98\ub9ac(NLP)\uc640 \ucd94\ucc9c \uc2dc\uc2a4\ud15c\uc5d0\uc11c \ub9ce\uc774 \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, \ub2e8\uc5b4\ub97c \ubca1\ud130\ub85c \ud45c\ud604\ud558\uc5ec \ub2e8\uc5b4 \uac04\uc758 \uc758\ubbf8 \uc720\uc0ac\uc131\uc744 \uacc4\uc0b0\ud560 \ub54c \uc784\ubca0\ub529 \uae30\ubc95\uc744 \uc0ac\uc6a9\ud569\ub2c8\ub2e4. \uc784\ubca0\ub529\uc740 \ub2e4\uc591\ud55c \ud615\ud0dc\ub85c \uc874\uc7ac\ud560 \uc218 \uc788\uc73c\uba70, \uc5ec\uae30\uc5d0\uc11c\ub294 \ub450 \uac00\uc9c0 \uc8fc\uc694 \ubc29\uc2dd\uc778 \ud69f\uc218 \uae30\ubc18 \uc784\ubca0\ub529\uacfc \uc608\uce21 \uae30\ubc18 \uc784\ubca0\ub529\uc5d0 \ub300\ud574 \uc124\uba85\ud569\ub2c8\ub2e4.\n <\/p>\n<\/section>\n

\n

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

\n \ud69f\uc218 \uae30\ubc18 \uc784\ubca0\ub529\uc740 \ud2b9\uc815 \ub370\uc774\ud130\uc758 \ubc1c\uc0dd \ube48\ub3c4 \uc218\ub97c \uae30\ubc18\uc73c\ub85c \uc784\ubca0\ub529\uc744 \uc218\ud589\ud558\ub294 \ubc29\uc2dd\uc785\ub2c8\ub2e4. \uac00\uc7a5 \ub300\ud45c\uc801\uc778 \uc608\ub85c\ub294 TF-IDF(vectorization), Bag of Words(BOW) \ub4f1\uc774 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ubc29\ubc95\ub4e4\uc740 \ub2e8\uc5b4\uc758 \ucd9c\ud604 \ube48\ub3c4\ub97c \uae30\ubc18\uc73c\ub85c \ud558\uc5ec \ubb38\uc11c\uc758 \ud2b9\uc9d5\uc744 \ud30c\uc545\ud569\ub2c8\ub2e4.\n <\/p>\n

2.1. TF-IDF \uc124\uba85<\/h3>\n

\n TF-IDF(Term Frequency-Inverse Document Frequency)\ub294 \ub2e8\uc5b4\uc758 \uc911\uc694\uc131\uc744 \ud3c9\uac00\ud558\uae30 \uc704\ud55c \ud1b5\uacc4\uc801 \uc218\uce58\uc785\ub2c8\ub2e4. TF\ub294 \ud2b9\uc815 \ub2e8\uc5b4\uac00 \ubb38\uc11c \ub0b4\uc5d0\uc11c \uc5bc\ub9c8\ub098 \uc790\uc8fc \ub098\ud0c0\ub098\ub294\uc9c0\ub97c \ub098\ud0c0\ub0b4\uba70, IDF\ub294 \ub300\ub7c9\uc758 \ubb38\uc11c \ub0b4\uc5d0\uc11c \ud2b9\uc815 \ub2e8\uc5b4\uac00 \uc5bc\ub9c8\ub098 \ub4dc\ubb3c\uac8c \ub098\ud0c0\ub098\ub294\uc9c0\ub97c \ub098\ud0c0\ub0c5\ub2c8\ub2e4.\n <\/p>\n

2.2. \ud30c\uc774\ud1a0\uce58\ub85c TF-IDF \uad6c\ud604\ud558\uae30<\/h3>\n

\uc544\ub798\ub294 \ud30c\uc774\ud1a0\uce58\ub97c \uc0ac\uc6a9\ud558\uc5ec \uac04\ub2e8\ud55c TF-IDF \uacc4\uc0b0 \uc608\uc81c\uc785\ub2c8\ub2e4.<\/p>\n

\nimport torch\nfrom sklearn.feature_extraction.text import TfidfVectorizer\nimport numpy as np\n\n# \uc0d8\ud50c \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\ndocuments = [\n    \"\uc774\uac83\uc740 \uccab \ubc88\uc9f8 \ubb38\uc11c\uc785\ub2c8\ub2e4.\",\n    \"\uc774 \ubb38\uc11c\ub294 \ub450 \ubc88\uc9f8 \ubb38\uc11c\uc785\ub2c8\ub2e4.\",\n    \"\uadf8\ub9ac\uace0 \uc774 \ubb38\uc11c\ub294 \uc138 \ubc88\uc9f8 \ubb38\uc11c\uc785\ub2c8\ub2e4.\",\n    \"\ubb38\uc11c\ub294 \uc5ec\uae30\uc11c \ub05d\ub0a9\ub2c8\ub2e4.\"\n]\n\n# TF-IDF \ubca1\ud130\ud654\nvectorizer = TfidfVectorizer()\ntfidf_matrix = vectorizer.fit_transform(documents)\ntfidf_array = tfidf_matrix.toarray()\n\n# \uacb0\uacfc \ucd9c\ub825\nprint(\"\ub2e8\uc5b4 \ubaa9\ub85d:\", vectorizer.get_feature_names_out())\nprint(\"TF-IDF \ud589\ub82c:\\n\", tfidf_array)\n        <\/code><\/pre>\n
\n            \uc704 \ucf54\ub4dc\ub294 TF-IDF \ubca1\ud130\ud654\ub97c \ud1b5\ud574 \uac01 \ubb38\uc11c\uc5d0\uc11c\uc758 \ub2e8\uc5b4 \ucd9c\ud604 \ube48\ub3c4\ub97c \uacc4\uc0b0\ud569\ub2c8\ub2e4. \uacb0\uacfc\uc801\uc73c\ub85c \ub2e8\uc5b4 \ubaa9\ub85d\uacfc \uac01 \ubb38\uc11c\uc5d0 \ub300\ud55c TF-IDF \ud589\ub82c\uc744 \ucd9c\ub825\ud569\ub2c8\ub2e4.\n        <\/p>\n<\/section>\n
\n
3. \uc608\uce21 \uae30\ubc18 \uc784\ubca0\ub529<\/h2>\n
\n            \uc608\uce21 \uae30\ubc18 \uc784\ubca0\ub529\uc740 \ub525\ub7ec\ub2dd \ubaa8\ub378\uc744 \ud1b5\ud574 \ub2e8\uc5b4 \ub610\ub294 \uc544\uc774\ud15c\uc758 \uc784\ubca0\ub529\uc744 \ud559\uc2b5\ud558\ub294 \ubc29\ubc95\uc785\ub2c8\ub2e4. Word2Vec, GloVe \ub4f1\uc758 \uae30\ubc95\uc774 \ub300\ud45c\uc801\uc785\ub2c8\ub2e4. \uc774 \ubc29\ubc95\uc740 \uc8fc\ubcc0 \ub2e8\uc5b4\ub97c \uae30\ubc18\uc73c\ub85c \ud2b9\uc815 \ub2e8\uc5b4\uc758 \uc784\ubca0\ub529\uc744 \ud559\uc2b5\ud558\ub294 \ubc29\uc2dd\uc774\uba70, \uc774 \uacb0\uacfc\ub85c \uc5bb\uc5b4\uc9c4 \uc784\ubca0\ub529\uc740 \uc758\ubbf8\uc801\uc73c\ub85c \uc720\uc0ac\ud55c \ub2e8\uc5b4 \uac04\uc758 \uac70\ub9ac\ub3c4 \uac00\uae4c\uc6cc\uc9c0\ub294 \ud2b9\uc131\uc744 \uac00\uc9d1\ub2c8\ub2e4.\n        <\/p>\n
3.1. Word2Vec \uc124\uba85<\/h3>\n
\n            Word2Vec\uc740 \ub2e8\uc5b4\ub97c \ubca1\ud130 \uacf5\uac04\uc5d0 \ub9e4\ud551\ud558\ub294 \ub300\ud45c\uc801\uc778 \uc608\uce21 \uae30\ubc18 \uc784\ubca0\ub529 \uae30\ubc95\uc73c\ub85c, Continuous Bag of Words (CBOW)\uc640 Skip-Gram \ub450 \uac00\uc9c0 \ubaa8\ub378\uc744 \uc81c\uacf5\ud569\ub2c8\ub2e4. CBOW \ubaa8\ub378\uc740 \uc8fc\uc5b4\uc9c4 \ub2e8\uc5b4\uc758 \uc8fc\ubcc0 \ub2e8\uc5b4\ub4e4\ub85c\ubd80\ud130 \uadf8 \ub2e8\uc5b4\ub97c \uc608\uce21\ud558\ub294 \ubc29\uc2dd\uc744 \uc0ac\uc6a9\ud558\uace0, Skip-Gram \ubaa8\ub378\uc740 \uc8fc\uc5b4\uc9c4 \ub2e8\uc5b4\ub85c\ubd80\ud130 \uc8fc\ubcc0 \ub2e8\uc5b4\ub4e4\uc744 \uc608\uce21\ud558\ub294 \ubc29\uc2dd\uc785\ub2c8\ub2e4.\n        <\/p>\n
3.2. PyTorch\ub85c Word2Vec \uad6c\ud604\ud558\uae30<\/h3>\n
\uc544\ub798\ub294 PyTorch\ub97c \uc0ac\uc6a9\ud558\uc5ec Skip-Gram \ubaa8\ub378\uc744 \uad6c\ud604\ud558\ub294 \uc608\uc81c\uc785\ub2c8\ub2e4.<\/p>\n
\nimport torch\nimport torch.nn as nn\nimport torch.optim as optim\nfrom collections import Counter\n\n# \uc0d8\ud50c \ub370\uc774\ud130\ub97c \uc2e4\ud589\ud560 \ud568\uc218 \uc815\uc758\ndef prepare_data(documents):\n    words = [word for doc in documents for word in doc.split()]\n    word_counts = Counter(words)\n    vocabulary_size = len(word_counts)\n    word2idx = {words: i for i, words in enumerate(word_counts.keys())}\n    idx2word = {i: words for words, i in word2idx.items()}\n    return word2idx, idx2word, vocabulary_size\n\n# Skip-Gram \ubaa8\ub378 \uc815\uc758\nclass SkipGramModel(nn.Module):\n    def __init__(self, vocab_size, embed_size):\n        super(SkipGramModel, self).__init__()\n        self.embedding = nn.Embedding(vocab_size, embed_size)\n\n    def forward(self, center_word):\n        return self.embedding(center_word)\n\n# \uc124\uc815 \ubc0f \ub370\uc774\ud130 \uc900\ube44\ndocuments = [\n    \"\uc774\uac83\uc740 \uccab \ubc88\uc9f8 \ubb38\uc11c\uc785\ub2c8\ub2e4\",\n    \"\uc774 \ubb38\uc11c\ub294 \ub450 \ubc88\uc9f8 \ubb38\uc11c\uc785\ub2c8\ub2e4\",\n    \"\uadf8\ub9ac\uace0 \uc774 \ubb38\uc11c\ub294 \uc138 \ubc88\uc9f8 \ubb38\uc11c\uc785\ub2c8\ub2e4\"\n]\nword2idx, idx2word, vocab_size = prepare_data(documents)\n\n# \ubaa8\ub378 \uc124\uc815 \ubc0f \ud6c8\ub828\nembed_size = 10\nmodel = SkipGramModel(vocab_size, embed_size)\nloss_function = nn.CrossEntropyLoss()\noptimizer = optim.SGD(model.parameters(), lr=0.01)\n\n# \uc608\uc81c \uc785\ub825\ninput_word = torch.tensor([word2idx['\uc774\uac83\uc740']])\ntarget_word = torch.tensor([word2idx['\uccab \ubc88\uc9f8']])\n\n# \ud6c8\ub828 \ud504\ub85c\uc138\uc2a4 (1 epochs \uc608\uc2dc)\nfor epoch in range(1):\n    model.zero_grad()\n    # \uc608\uce21\n    predictions = model(input_word)\n    # \uc190\uc2e4 \uacc4\uc0b0\n    loss = loss_function(predictions.view(1, -1), target_word)\n    loss.backward()\n    optimizer.step()\n    \n# \uacb0\uacfc \ucd9c\ub825\nprint(\"\ub2e8\uc5b4 '\uc774\uac83\uc740'\uc758 \uc784\ubca0\ub529 \ubca1\ud130:\\n\", model.embedding.weight[word2idx['\uc774\uac83\uc740']].detach().numpy())\n        <\/code><\/pre>\n
\n            \uc704 \ucf54\ub4dc\ub294 PyTorch\ub97c \uc0ac\uc6a9\ud558\uc5ec Skip-Gram \ubaa8\ub378\uc744 \uac04\ub2e8\ud788 \uad6c\ud604\ud55c \uac83\uc785\ub2c8\ub2e4. \uac01 \ub2e8\uc5b4 \uc5e0\ubca0\ub529\uc744 \ud559\uc2b5\ud558\uace0, \ud2b9\uc815 \ub2e8\uc5b4\uc758 \uc784\ubca0\ub529 \ubca1\ud130\ub97c \ucd9c\ub825\ud569\ub2c8\ub2e4.\n        <\/p>\n<\/section>\n
\n
4. \uacb0\ub860<\/h2>\n
\n            \uc774\ubc88 \uae00\uc5d0\uc11c\ub294 \uc784\ubca0\ub529\uc758 \uac1c\ub150\uacfc \ud568\uaed8 \ud69f\uc218 \uae30\ubc18 \ubc0f \uc608\uce21 \uae30\ubc18 \uc784\ubca0\ub529 \uae30\ubc95\uc5d0 \ub300\ud574 \uc54c\uc544\ubcf4\uc558\uc2b5\ub2c8\ub2e4. TF-IDF\uc640 \uac19\uc740 \ud69f\uc218 \uae30\ubc18 \ubc29\uc2dd\uc740 \ub370\uc774\ud130\uc758 \ubc1c\uc0dd \ube48\ub3c4\ub97c \ubc14\ud0d5\uc73c\ub85c \ud55c \uac83\uc774\uace0, Word2Vec\uacfc \uac19\uc740 \uc608\uce21 \uae30\ubc18 \ubc29\uc2dd\uc740 \ub525\ub7ec\ub2dd \ubaa8\ub378\uc744 \ud1b5\ud574 \ub2e8\uc5b4\uc758 \uc758\ubbf8\ub97c \ud559\uc2b5\ud558\ub294 \ubc29\uc2dd\uc785\ub2c8\ub2e4. \uac01 \uc784\ubca0\ub529 \uae30\ubc95\uc758 \ud2b9\uc131\uc744 \uc774\ud574\ud558\uace0, \uc2e4\uc81c \uc608\uc81c\ub97c \ud1b5\ud574 \uc774\ub97c \uc801\uc6a9\ud558\ub294 \uacfc\uc815\uc744 \ubc30\uc6e0\uc2b5\ub2c8\ub2e4.\n        <\/p>\n
\n            \ub525\ub7ec\ub2dd\uc5d0\uc11c\ub294 \ub370\uc774\ud130\uc758 \ud2b9\uc131\uc744 \uc774\ud574\ud558\uace0 \uc774\ub97c \uae30\ubc18\uc73c\ub85c \ud55c \uc784\ubca0\ub529 \uae30\ubc95\uc758 \uc120\ud0dd\uc774 \ub9e4\uc6b0 \uc911\uc694\ud55c\ub370, \uc774\ub97c \ud1b5\ud574 \ubaa8\ub378\uc758 \uc131\ub2a5\uc744 \ud06c\uac8c \ud5a5\uc0c1\uc2dc\ud0ac \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ub2e4\uc74c \ub0b4\uc6a9\uc5d0\uc11c\ub294 \uc774\ub4e4 \uae30\ubc95\uc744 \ud655\uc7a5\ud558\uc5ec \ub354 \ubcf5\uc7a1\ud55c \ubaa8\ub378\uc744 \uad6c\ud604\ud574 \ub098\uac00\ub294 \ubc29\ubc95\uc744 \ub2e4\ub8f0 \uc608\uc815\uc774\ub2c8 \ub9ce\uc740 \uad00\uc2ec \ubd80\ud0c1\ub4dc\ub9bd\ub2c8\ub2e4.\n        <\/p>\n<\/section>\n