12. Sequence Labeling/12-4. named_entity_recognition_using_bilstm.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Named Entity Recognition using BiLSTM.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "source": [
        "이 자료는 위키독스 딥 러닝을 이용한 자연어 처리 입문의 BiLSTM을 이용한 개체명 인식의 학습 자료입니다.  \n",
        "\n",
        "링크 : https://wikidocs.net/24682"
      ],
      "metadata": {
        "id": "nOz2C4cWn5sL"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import urllib.request"
      ],
      "metadata": {
        "id": "_lolJq4xltUQ"
      },
      "execution_count": 1,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "urllib.request.urlretrieve(\"https://raw.githubusercontent.com/ukairia777/tensorflow-nlp-tutorial/main/12.%20Sequence%20Labeling/dataset/train.txt\", filename=\"train.txt\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wNQa4_UPluZ5",
        "outputId": "f110fabf-77ed-49e2-a1c0-53dbebfdf628"
      },
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "('train.txt', <http.client.HTTPMessage at 0x7f7733f65890>)"
            ]
          },
          "metadata": {},
          "execution_count": 2
        }
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "id": "vh1vovqEligf"
      },
      "outputs": [],
      "source": [
        "import re\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "from tensorflow.keras.preprocessing.text import Tokenizer\n",
        "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
        "from tensorflow.keras.utils import to_categorical\n",
        "from sklearn.model_selection import train_test_split"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "f = open('train.txt', 'r')\n",
        "tagged_sentences = []\n",
        "sentence = []\n",
        "\n",
        "for line in f:\n",
        "    if len(line)==0 or line.startswith('-DOCSTART') or line[0]==\"\\n\":\n",
        "        if len(sentence) > 0:\n",
        "            tagged_sentences.append(sentence)\n",
        "            sentence = []\n",
        "        continue\n",
        "    splits = line.split(' ') # 공백을 기준으로 속성을 구분한다.\n",
        "    splits[-1] = re.sub(r'\\n', '', splits[-1]) # 줄바꿈 표시 \\n을 제거한다.\n",
        "    word = splits[0].lower() # 단어들은 소문자로 바꿔서 저장한다.\n",
        "    sentence.append([word, splits[-1]]) # 단어와 개체명 태깅만 기록한다."
      ],
      "metadata": {
        "id": "5nKjvwc7lqpx"
      },
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(\"전체 샘플 개수: \", len(tagged_sentences)) # 전체 샘플의 개수 출력"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "HJrXmODcl6Rq",
        "outputId": "93689f2a-913c-4b26-af3e-beaef0b543a7"
      },
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "전체 샘플 개수:  14041\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print(tagged_sentences[0]) # 첫번째 샘플 출력"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "zYoraoi4mVeB",
        "outputId": "ded0bb27-b7a1-446e-fd29-6d150389b26c"
      },
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[['eu', 'B-ORG'], ['rejects', 'O'], ['german', 'B-MISC'], ['call', 'O'], ['to', 'O'], ['boycott', 'O'], ['british', 'B-MISC'], ['lamb', 'O'], ['.', 'O']]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "sentences, ner_tags = [], [] \n",
        "for tagged_sentence in tagged_sentences: # 14,041개의 문장 샘플을 1개씩 불러온다.\n",
        "    sentence, tag_info = zip(*tagged_sentence) # 각 샘플에서 단어들은 sentence에 개체명 태깅 정보들은 tag_info에 저장.\n",
        "    sentences.append(list(sentence)) # 각 샘플에서 단어 정보만 저장한다.\n",
        "    ner_tags.append(list(tag_info)) # 각 샘플에서 개체명 태깅 정보만 저장한다."
      ],
      "metadata": {
        "id": "5I7QkSPdmWAU"
      },
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(sentences[0])\n",
        "print(ner_tags[0])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "5v_-xs4CmaXx",
        "outputId": "999a91f1-b11b-4322-99ce-7a2cac5610ab"
      },
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "['eu', 'rejects', 'german', 'call', 'to', 'boycott', 'british', 'lamb', '.']\n",
            "['B-ORG', 'O', 'B-MISC', 'O', 'O', 'O', 'B-MISC', 'O', 'O']\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print(sentences[12])\n",
        "print(ner_tags[12])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "gXM4arC2mmnx",
        "outputId": "32fb0fa4-245e-47b1-f3cd-ebb36e7ef3b6"
      },
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "['only', 'france', 'and', 'britain', 'backed', 'fischler', \"'s\", 'proposal', '.']\n",
            "['O', 'B-LOC', 'O', 'B-LOC', 'O', 'B-PER', 'O', 'O', 'O']\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print('샘플의 최대 길이 : %d' % max(len(l) for l in sentences))\n",
        "print('샘플의 평균 길이 : %f' % (sum(map(len, sentences))/len(sentences)))\n",
        "plt.hist([len(s) for s in sentences], bins=50)\n",
        "plt.xlabel('length of samples')\n",
        "plt.ylabel('number of samples')\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 315
        },
        "id": "RQ8_v9WMmnpC",
        "outputId": "973d5110-3535-4da7-fd83-96801cbbe2b9"
      },
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "샘플의 최대 길이 : 113\n",
            "샘플의 평균 길이 : 14.501887\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYsAAAEGCAYAAACUzrmNAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAcT0lEQVR4nO3de5QdZZnv8e/PyE0EE0zLCkm0gwYUHQmhubgED4pAuIzAOQrkjIKARBQGHNExoAcQhyWMCMowJxogk+DhIktAciQCkeEyHrmkAzlJuC0ChEMyIWkEEi4aSPKcP+rduul0d1U6u/atf5+19tpVT92ebUk/qaq33lcRgZmZ2UDe0egEzMys+blYmJlZLhcLMzPL5WJhZma5XCzMzCzXOxudQFlGjhwZnZ2djU7DzKxlzJ8//8WI6OhrWdsWi87OTrq7uxudhplZy5D0XH/LSrsNJWmspLslPSbpUUlnpvgOkuZKeip9j0hxSbpc0hJJCyVNrNrXCWn9pySdUFbOZmbWtzKfWawDzoqI3YB9gdMk7QZMBe6KiPHAXWke4FBgfPpMAaZBVlyA84B9gL2B8yoFxszM6qO0YhERKyLi4TT9KvA4MBo4EpiVVpsFHJWmjwSuicwDwHBJo4BDgLkR8VJEvAzMBSaVlbeZmW2sLq2hJHUCewAPAjtGxIq06AVgxzQ9Gni+arNlKdZfvK/jTJHULam7p6enZvmbmQ11pRcLSe8GbgK+ERFrqpdF1jFVzTqniojpEdEVEV0dHX0+0Dczs0EotVhI2oKsUFwbETen8Mp0e4n0vSrFlwNjqzYfk2L9xc3MrE7KbA0l4Grg8Yi4tGrRbKDSoukE4Naq+PGpVdS+wOp0u+oO4GBJI9KD7YNTzMzM6qTM9yw+CXwJWCRpQYqdA1wE3CjpZOA54Ji0bA5wGLAEeAM4ESAiXpL0A2BeWu+CiHipxLzNzKwXtet4Fl1dXeGX8szMipM0PyK6+lrWtm9wN4POqbf1GV960eF1zsTMbPO4I0EzM8vlYmFmZrlcLMzMLJeLhZmZ5XKxMDOzXG4N1Qe3YjIzeztfWZiZWS4XCzMzy+ViYWZmuVwszMwsl4uFmZnlcrEwM7NcLhZmZpbLxcLMzHK5WJiZWS4XCzMzy+ViYWZmuUorFpJmSFolaXFV7JeSFqTP0srY3JI6Jf2patnPqrbZU9IiSUskXS5JZeVsZmZ9K7MjwZnAFcA1lUBEHFuZlvRjYHXV+k9HxIQ+9jMNOAV4EJgDTAJ+W0K+ZmbWj9KuLCLiPuClvpalq4NjgOsH2oekUcD2EfFARARZ4Tmq1rmamdnAGvXMYn9gZUQ8VRUbJ+kRSfdK2j/FRgPLqtZZlmJ9kjRFUrek7p6entpnbWY2RDWqWEzm7VcVK4D3R8QewDeB6yRtv6k7jYjpEdEVEV0dHR01StXMzOo++JGkdwL/FdizEouItcDaND1f0tPALsByYEzV5mNSzMzM6qgRVxafBZ6IiL/cXpLUIWlYmt4ZGA88ExErgDWS9k3POY4Hbm1AzmZmQ1qZTWevB+4HdpW0TNLJadFxbPxg+1PAwtSU9lfAqRFReTj+deAqYAnwNG4JZWZWd6XdhoqIyf3Ev9xH7Cbgpn7W7wY+VtPkzMxsk/gNbjMzy+ViYWZmuVwszMwsl4uFmZnlcrEwM7NcLhZmZpbLxcLMzHK5WJiZWS4XCzMzy+ViYWZmuVwszMwsl4uFmZnlcrEwM7NcLhZmZpbLxcLMzHK5WJiZWS4XCzMzy1XaSHmSZgBHAKsi4mMpdj5wCtCTVjsnIuakZWcDJwPrgTMi4o4UnwT8FBgGXBURF5WV82B1Tr2t0SmYmZWqtGIBzASuAK7pFb8sIi6pDkjajWxs7o8COwG/k7RLWvyvwEHAMmCepNkR8ViJeffLRcHMhqoyx+C+T1JnwdWPBG6IiLXAs5KWAHunZUsi4hkASTekdRtSLMzMhqpGPLM4XdJCSTMkjUix0cDzVessS7H+4mZmVkf1LhbTgA8CE4AVwI9ruXNJUyR1S+ru6enJ38DMzAqpa7GIiJURsT4iNgBX8tdbTcuBsVWrjkmx/uL97X96RHRFRFdHR0dtkzczG8LqWiwkjaqaPRpYnKZnA8dJ2krSOGA88BAwDxgvaZykLckegs+uZ85mZlZu09nrgQOAkZKWAecBB0iaAASwFPgqQEQ8KulGsgfX64DTImJ92s/pwB1kTWdnRMSjZeVsZmZ9yy0Wkr4A3B4Rr0r6HjAR+KeIeHig7SJich/hqwdY/0Lgwj7ic4A5eXmamVl5ityG+h+pUOwHfJbsD/60ctMyM7NmUqRYrE/fhwPTI+I2YMvyUjIzs2ZTpFgsl/Rz4FhgjqStCm5nZmZtosgf/WPIHjAfEhGvADsA3y41KzMzayq5xSIi3gBWAful0DrgqTKTMjOz5pJbLCSdB3wHODuFtgD+V5lJmZlZcylyG+po4HPA6wAR8Z/AdmUmZWZmzaVIsXgzIoLsRTokbVtuSmZm1myKFIsbU2uo4ZJOAX5H1q+TmZkNEblvcEfEJZIOAtYAuwLnRsTc0jMzM7OmUahvqFQcXCDMzIaofouFpFdJzyl6LwIiIrYvLSszM2sq/RaLiHCLJzMzAwrehpI0keylvAB+HxGPlJqVmZk1lSIv5Z0LzALeC4wEZqauys3MbIgocmXxd8DuEfFnAEkXAQuAfyozMTMzax5F3rP4T2DrqvmtGGAcbDMzaz9FrixWA49Kmkv2zOIg4CFJlwNExBkl5mdmZk2gSLG4JX0q7imyY0kzgCOAVRHxsRT7EfC3wJvA08CJEfGKpE7gceDJtPkDEXFq2mZPYCawDdnwqmem7kfMzKxOirzBPWuQ+54JXAFcUxWbC5wdEeskXUzWk+130rKnI2JCH/uZBpwCPEhWLCYBvx1kTmZmNghFWkMdIekRSS9JWiPpVUlr8raLiPuAl3rF7oyIdWn2AWBMzrFHAdtHxAPpauIa4Ki8Y5uZWW0VecD9E+AE4L0RsX1EbFejt7dP4u1XCONSUbpX0v4pNhpYVrXOshTrk6Qpkroldff09NQgRTMzg2LF4nlgcS2fE0j6LtmIe9em0Arg/RGxB/BN4DpJm1yQImJ6RHRFRFdHR0et0jUzG/KKPOD+R2COpHuBtZVgRFw6mANK+jLZg+8DKwUoItZW9h0R8yU9DexC1kS3+lbVGNqg2W7n1Nv6jC+96PA6Z2JmVkyRK4sLgTfI3rXYruqzySRNIis+n0tje1fiHZKGpemdgfHAMxGxAlgjaV9JAo4Hbh3Msc3MbPCKXFnsVGn6uikkXQ8cAIyUtAw4j6z101bA3Oxv/1+ayH4KuEDSW8AG4NSIqDwc/zp/bTr7W9wSysys7ooUizmSDo6IOzdlxxExuY/w1f2sexNwUz/LuoFNLlZmZlY7RW5DfQ24XdKfNqXprJmZtY8iL+V5XAszsyGu6HgWI8geOv+lQ8H00p2ZmQ0BucVC0leAM8marS4A9gXuBz5TbmpmZtYsijyzOBPYC3guIj4N7AG8UmpWZmbWVIoUiz9XDXy0VUQ8AexablpmZtZMijyzWCZpOPBrsvcjXgaeKzctMzNrJkVaQx2dJs+XdDfwHuD2UrMyM7OmUqSL8g9K2qoyC3QC7yozKTMzay5FnlncBKyX9CFgOjAWuK7UrMzMrKkUKRYb0oBFRwP/EhHfBkaVm5aZmTWTIsXiLUmTyQZA+k2KbVFeSmZm1myKFIsTgU8AF0bEs5LGAb8oNy0zM2smRVpDPQacUTX/LHBxmUmZmVlzKXJlYWZmQ5yLhZmZ5eq3WEj6Rfo+s37pmJlZMxroymJPSTsBJ0kaIWmH6k+RnUuaIWmVpMVVsR0kzZX0VPoekeKSdLmkJZIWSppYtc0Jaf2nJJ0w2B9rZmaDM1Cx+BlwF/BhYH6vT3fB/c8EJvWKTQXuiojxaf9TU/xQsjEzxgNTgGmQFRey8bv3AfYGzqsUGDMzq49+i0VEXB4RHwFmRMTOETGu6rNzkZ2nAZJe6hU+EpiVpmcBR1XFr4nMA8BwSaOAQ4C5EfFSRLwMzGXjAmRmZiUq0nT2a5J2B/ZPofsiYuFmHHPHiFiRpl8AdkzTo4Hnq9ZblmL9xc3MrE6KdCR4BnAt8L70uVbS39fi4BERQNRiXwCSpkjqltTd09NTq92amQ15RZrOfgXYJyLOjYhzyYZVPWUzjrky3V4ifa9K8eVknRRWjEmx/uIbiYjpEdEVEV0dHR2bkaKZmVUrUiwErK+aX59igzWbrJ8p0vetVfHjU6uofYHV6XbVHcDBqUXWCODgFDMzszopMlLevwEPSrolzR8FXF1k55KuBw4ARkpaRtaq6SLgRkknk424d0xafQ5wGLAEeIOsTyoi4iVJPwDmpfUuiIjeD83NzKxERR5wXyrpHmC/FDoxIh4psvOImNzPogP7WDeA0/rZzwxgRpFjmplZ7RW5siAiHgYeLjkXMzNrUu4byszMcrlYmJlZrgGLhaRhku6uVzJmZtacBiwWEbEe2CDpPXXKx8zMmlCRB9yvAYskzQVerwQj4oz+NzEzs3ZSpFjcnD5mZjZEFXnPYpakbYD3R8STdcjJzMyaTJGOBP8WWADcnuYnSJpddmJmZtY8ijSdPZ9s0KFXACJiAVBoPAszM2sPRYrFWxGxuldsQxnJmJlZcyrygPtRSf8dGCZpPHAG8Idy0zIzs2ZS5Mri74GPAmuB64E1wDfKTMrMzJpLkdZQbwDflXRxNhuvlp+WmZk1kyKtofaStAhYSPZy3v+VtGf5qZmZWbMo8sziauDrEfEfAJL2IxsQ6eNlJmZmZs2jyDOL9ZVCARARvwfWlZeSmZk1m36vLCRNTJP3Svo52cPtAI4F7ik/NTMzaxYD3Yb6ca/586qmY7AHlLQr8Muq0M7AucBw4BSgJ8XPiYg5aZuzgZOB9cAZEXHHYI9vZmabrt9iERGfLuOAqX+pCZCNlwEsB24BTgQui4hLqteXtBtwHFnz3Z2A30naJXWfbmZmdZD7gFvScOB4oLN6/Rp1UX4g8HREPCepv3WOBG6IiLXAs5KWkHU/cn8Njm9mZgUUecA9h6xQLALmV31q4TiyZyEVp0taKGmGpBEpNhp4vmqdZSm2EUlTJHVL6u7p6elrFTMzG4QiTWe3johv1vrAkrYEPgecnULTgB+QPQ/5Adkzk5M2ZZ8RMR2YDtDV1TXo5ypmZvZ2Ra4sfiHpFEmjJO1Q+dTg2IcCD0fESoCIWBkR6yNiA3Al2a0myJ5pjK3abkyKmZlZnRQpFm8CPyJ7RlC5BdVdg2NPpuoWlKRRVcuOBhan6dnAcZK2kjQOGA88VIPjm5lZQUVuQ50FfCgiXqzVQSVtCxwEfLUq/M+SJpDdhlpaWRYRj0q6EXiM7GXA09wSysysvooUiyXAG7U8aES8Dry3V+xLA6x/IXBhLXMwM7PiihSL14EFku4m66YcqFnTWTMzawFFisWv08fMzIaoIuNZzKpHImZm1ryKvMH9LH30BRURO5eSkZmZNZ0it6G6qqa3Br4A1OI9CzMzaxFFbkP9sVfoJ5Lmk/UUazXUOfW2PuNLLzq8zpmYmb1dkdtQE6tm30F2pVHkisTMzNpEkT/61eNarCN7Ye6YUrKxluGrILOhpchtqFLGtbDy9PeHHPzH3MwGp8htqK2A/8bG41lcUF5aZmbWTIrchroVWE3WgeDanHXNzKwNFSkWYyJiUumZmJlZ0yrSRfkfJP1N6ZmYmVnTKnJlsR/w5fQm91pAQETEx0vNzMzMmkaRYnFo6VmYmVlTK9J09rl6JGJmZs2ryDMLMzMb4lwszMwsV8OKhaSlkhZJWiCpO8V2kDRX0lPpe0SKS9LlkpZIWtirvyozMytZozsE/HREvFg1PxW4KyIukjQ1zX+H7CH7+PTZB5iWvocE98NkZo3WbLehjgQqI/PNAo6qil8TmQeA4ZJGNSJBM7OhqJHFIoA7Jc2XNCXFdoyIFWn6BWDHND0aeL5q22Up9jaSpkjqltTd09NTVt5mZkNOI29D7RcRyyW9D5gr6YnqhRERkjYaznUgETEdmA7Q1dW1SduamVn/GnZlERHL0/cq4BZgb2Bl5fZS+l6VVl8OjK3afEyKmZlZHTSkWEjaVtJ2lWngYGAxMBs4Ia12AlmPt6T48alV1L7A6qrbVWZmVrJG3YbaEbhFUiWH6yLidknzgBslnQw8x19H5JsDHAYsAd4ATqx/ymZmQ1dDikVEPAPs3kf8j8CBfcQDOK0OqZmZWR8a/Z6FNQm/y2FmA3GxaGEDjbVdy23MzJrtpTwzM2tCvrKwAflKxMzAVxZmZlaAryyspvyg3Kw9+crCzMxyuViYmVkuFwszM8vlYmFmZrlcLMzMLJeLhZmZ5XLTWasLN6k1a22+sjAzs1y+srCm5CsRs+biKwszM8vlYmFmZrnqXiwkjZV0t6THJD0q6cwUP1/SckkL0uewqm3OlrRE0pOSDql3zmZmQ10jnlmsA86KiIclbQfMlzQ3LbssIi6pXlnSbsBxwEeBnYDfSdolItbXNWszsyGs7sUiIlYAK9L0q5IeB0YPsMmRwA0RsRZ4VtISYG/g/tKTtdJ5vAyz1tDQZxaSOoE9gAdT6HRJCyXNkDQixUYDz1dttox+ioukKZK6JXX39PSUlLWZ2dDTsGIh6d3ATcA3ImINMA34IDCB7Mrjx5u6z4iYHhFdEdHV0dFR03zNzIayhhQLSVuQFYprI+JmgIhYGRHrI2IDcCXZrSaA5cDYqs3HpJiZmdVJI1pDCbgaeDwiLq2Kj6pa7WhgcZqeDRwnaStJ44DxwEP1ytfMzBrTGuqTwJeARZIWpNg5wGRJE4AAlgJfBYiIRyXdCDxG1pLqNLeEMjOrr0a0hvo9oD4WzRlgmwuBC0tLyszMBuQ3uM3MLJeLhZmZ5XKxMDOzXC4WZmaWy8XCzMxyefAjaykeFMmsMXxlYWZmuVwszMwsl29DWVvw7SmzcvnKwszMcrlYmJlZLhcLMzPL5WcW1tYGGrbVzzPMivOVhZmZ5fKVhQ1ZbkFlVpyvLMzMLJeLhZmZ5XKxMDOzXC3zzELSJOCnwDDgqoi4qMEpWZvyswyzjbVEsZA0DPhX4CBgGTBP0uyIeKyxmdlQ4iJiQ1lLFAtgb2BJRDwDIOkG4EjAxcIabqB3OTaFi441s1YpFqOB56vmlwH79F5J0hRgSpp9TdKTm3CMkcCLg86webXr74I2+226+C+TbfW7qrTr74L2+W0f6G9BqxSLQiJiOjB9MNtK6o6Irhqn1HDt+rugfX+bf1fraeffVtEqraGWA2Or5sekmJmZ1UGrFIt5wHhJ4yRtCRwHzG5wTmZmQ0ZL3IaKiHWSTgfuIGs6OyMiHq3xYQZ1+6oFtOvvgvb9bf5draedfxsAiohG52BmZk2uVW5DmZlZA7lYmJlZLhcLsq5EJD0paYmkqY3OZ7AkjZV0t6THJD0q6cwU30HSXElPpe8Rjc51MCQNk/SIpN+k+XGSHkzn7Zep8UPLkTRc0q8kPSHpcUmfaIdzJukf0v8PF0u6XtLWrXrOJM2QtErS4qpYn+dImcvTb1woaWLjMq+dIV8sqroSORTYDZgsabfGZjVo64CzImI3YF/gtPRbpgJ3RcR44K4034rOBB6vmr8YuCwiPgS8DJzckKw230+B2yPiw8DuZL+xpc+ZpNHAGUBXRHyMrGHKcbTuOZsJTOoV6+8cHQqMT58pwLQ65ViqIV8sqOpKJCLeBCpdibSciFgREQ+n6VfJ/uiMJvs9s9Jqs4CjGpPh4EkaAxwOXJXmBXwG+FVapVV/13uATwFXA0TEmxHxCm1wzshaW24j6Z3Au4AVtOg5i4j7gJd6hfs7R0cC10TmAWC4pFH1ybQ8LhZ9dyUyukG51IykTmAP4EFgx4hYkRa9AOzYoLQ2x0+AfwQ2pPn3Aq9ExLo036rnbRzQA/xbusV2laRtafFzFhHLgUuA/0dWJFYD82mPc1bR3zlqy78pLhZtSNK7gZuAb0TEmuplkbWVbqn20pKOAFZFxPxG51KCdwITgWkRsQfwOr1uObXoORtB9i/sccBOwLZsfBunbbTiOdpULhZt1pWIpC3ICsW1EXFzCq+sXAan71WNym+QPgl8TtJSstuEnyG7zz883eKA1j1vy4BlEfFgmv8VWfFo9XP2WeDZiOiJiLeAm8nOYzucs4r+zlFb/U2pcLFoo65E0n38q4HHI+LSqkWzgRPS9AnArfXObXNExNkRMSYiOsnOz79HxN8BdwOfT6u13O8CiIgXgOcl7ZpCB5J1vd/S54zs9tO+kt6V/n9Z+V0tf86q9HeOZgPHp1ZR+wKrq25XtSy/wQ1IOozsnnilK5ELG5zSoEjaD/gPYBF/vbd/DtlzixuB9wPPAcdERO+HdS1B0gHAtyLiCEk7k11p7AA8AnwxItY2Mr/BkDSB7MH9lsAzwIlk/5Br6XMm6fvAsWSt9B4BvkJ2777lzpmk64EDyLoiXwmcB/yaPs5RKo5XkN12ewM4MSK6G5F3LblYmJlZLt+GMjOzXC4WZmaWy8XCzMxyuViYmVkuFwszM8vlYmEtT9JrJexzQmpSXZk/X9K3NmN/X0g9yt5dmwwHncdSSSMbmYO1JhcLs75NAA7LXau4k4FTIuLTNdynWd24WFhbkfRtSfPSOALfT7HO9K/6K9P4CndK2iYt2yutu0DSj9LYC1sCFwDHpvixafe7SbpH0jOSzujn+JMlLUr7uTjFzgX2A66W9KNe64+SdF86zmJJ+6f4NEndKd/vV62/VNIP0/rdkiZKukPS05JOTesckPZ5m7JxWn4maaP/1iV9UdJDaV8/VzZeyDBJM1MuiyT9w2aeEmsXEeGPPy39AV5L3wcD0wGR/UPoN2Tdf3eSvUU8Ia13I9mbwwCLgU+k6YuAxWn6y8AVVcc4H/gDsBXZW7x/BLbolcdOZN1cdJB1EPjvwFFp2T1kYzv0zv0s4LtpehiwXZreoSp2D/DxNL8U+FqavgxYCGyXjrkyxQ8A/gzsnLafC3y+avuRwEeA/135DcD/BI4H9gTmVuU3vNHn15/m+PjKwtrJwenzCPAw8GGyAWgg69RuQZqeD3RKGk72x/n+FL8uZ/+3RcTaiHiRrNO43t2G7wXcE1nneeuAa8mK1UDmASdKOh/4m8jGIQE4RtLD6bd8lGxgropK32WLgAcj4tWI6AHWpt8E8FBkY7SsB64nu7KpdiBZYZgnaUGa35msu5GdJf2LpEnAGszI/vVj1i4E/DAifv62YDa2R3X/Q+uBbQax/9772Oz/fiLiPkmfIhvYaaakS8n69/oWsFdEvCxpJrB1H3ls6JXThqqcevfj03tewKyIOLt3TpJ2Bw4BTgWOAU7a1N9l7cdXFtZO7gBOSuN5IGm0pPf1t3JkI9K9KmmfFDquavGrZLd3NsVDwH+RNFLZcL2TgXsH2kDSB8huH11J1pngRGB7snEtVkvakWyYzk21d+pJ+R1knfn9vtfyu4DPV/73UTae9AdSS6l3RMRNwPdSPma+srD2ERF3SvoIcH/W8SevAV8kuwroz8nAlZI2kP1hX53idwNT0y2aHxY8/gpJU9O2IrttldcF9wHAtyW9lfI9PiKelfQI8ATZiGv/p8jxe5lH1vPph1I+t/TK9TFJ3wPuTAXlLeA04E9ko/ZV/iG50ZWHDU3uddaGNEnvjojX0vRUYFREnNngtDZLdTfujc7F2oevLGyoO1zS2WT/LTxH1grKzHrxlYWZmeXyA24zM8vlYmFmZrlcLMzMLJeLhZmZ5XKxMDOzXP8f2R6ZJNtq/9wAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "max_words = 4000\n",
        "\n",
        "src_tokenizer = Tokenizer(num_words=max_words, oov_token='OOV')\n",
        "src_tokenizer.fit_on_texts(sentences)\n",
        "tar_tokenizer = Tokenizer()\n",
        "tar_tokenizer.fit_on_texts(ner_tags)"
      ],
      "metadata": {
        "id": "dy9Gecafmovp"
      },
      "execution_count": 14,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "vocab_size = max_words\n",
        "tag_size = len(tar_tokenizer.word_index) + 1\n",
        "print('단어 집합의 크기 : {}'.format(vocab_size))\n",
        "print('개체명 태깅 정보 집합의 크기 : {}'.format(tag_size))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "IjPs5nEcnWP2",
        "outputId": "b7d34e75-3430-4f56-d92e-b8975990173e"
      },
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "단어 집합의 크기 : 4000\n",
            "개체명 태깅 정보 집합의 크기 : 10\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "X_train = src_tokenizer.texts_to_sequences(sentences)\n",
        "y_train = tar_tokenizer.texts_to_sequences(ner_tags)"
      ],
      "metadata": {
        "id": "YDekoU7_nXoN"
      },
      "execution_count": 16,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print(X_train[0])\n",
        "print(y_train[0])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "7zazi9ZqnYm8",
        "outputId": "38854b5e-92a4-457b-d600-7b1543118556"
      },
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[989, 1, 205, 629, 7, 3939, 216, 1, 3]\n",
            "[4, 1, 7, 1, 1, 1, 7, 1, 1]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "index_to_word = src_tokenizer.index_word\n",
        "index_to_ner = tar_tokenizer.index_word"
      ],
      "metadata": {
        "id": "Gh8UuEn4nZcE"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "decoded = []\n",
        "for index in X_train[0] : # 첫번째 샘플 안의 인덱스들에 대해서\n",
        "    decoded.append(index_to_word[index]) # 다시 단어로 변환\n",
        "\n",
        "print('기존 문장 : {}'.format(sentences[0]))\n",
        "print('빈도수가 낮은 단어가 OOV 처리된 문장 : {}'.format(decoded))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "q2c2ezsinaTl",
        "outputId": "263257f2-56d3-4f40-bdb1-82bd2a57bac2"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "기존 문장 : ['eu', 'rejects', 'german', 'call', 'to', 'boycott', 'british', 'lamb', '.']\n",
            "빈도수가 낮은 단어가 OOV 처리된 문장 : ['eu', 'OOV', 'german', 'call', 'to', 'boycott', 'british', 'OOV', '.']\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "max_len = 70\n",
        "\n",
        "X_train = pad_sequences(X_train, padding='post', maxlen=max_len)\n",
        "y_train = pad_sequences(y_train, padding='post', maxlen=max_len)"
      ],
      "metadata": {
        "id": "bP-d4XOonbPN"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "X_train, X_test, y_train, y_test = train_test_split(X_train, y_train, test_size=.2, random_state=777)"
      ],
      "metadata": {
        "id": "ZVLKCFZNndkT"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "y_train = to_categorical(y_train, num_classes=tag_size)\n",
        "y_test = to_categorical(y_test, num_classes=tag_size)"
      ],
      "metadata": {
        "id": "aNhwuVAynefj"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "print('훈련 샘플 문장의 크기 : {}'.format(X_train.shape))\n",
        "print('훈련 샘플 레이블의 크기 : {}'.format(y_train.shape))\n",
        "print('테스트 샘플 문장의 크기 : {}'.format(X_test.shape))\n",
        "print('테스트 샘플 레이블의 크기 : {}'.format(y_test.shape))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "OJjenkytnfe0",
        "outputId": "cc137b46-3d19-45d0-f507-a8c07e8a2440"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "훈련 샘플 문장의 크기 : (11232, 70)\n",
            "훈련 샘플 레이블의 크기 : (11232, 70, 10)\n",
            "테스트 샘플 문장의 크기 : (2809, 70)\n",
            "테스트 샘플 레이블의 크기 : (2809, 70, 10)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from tensorflow.keras.models import Sequential\n",
        "from tensorflow.keras.layers import Dense, Embedding, LSTM, Bidirectional, TimeDistributed\n",
        "from tensorflow.keras.optimizers import Adam"
      ],
      "metadata": {
        "id": "-BXzNIGCngX8"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model = Sequential()\n",
        "model.add(Embedding(input_dim=vocab_size, output_dim=128, input_length=max_len, mask_zero=True))\n",
        "model.add(Bidirectional(LSTM(256, return_sequences=True)))\n",
        "model.add(TimeDistributed(Dense(tag_size, activation='softmax')))"
      ],
      "metadata": {
        "id": "_mvoBvq-nqmc"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model.compile(loss='categorical_crossentropy', optimizer=Adam(0.001), metrics=['accuracy'])"
      ],
      "metadata": {
        "id": "iU3ZzE4Enr_0"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model.fit(X_train, y_train, batch_size=128, epochs=8,  validation_data=(X_test, y_test))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "TVAnH0cVntY0",
        "outputId": "8ad4373c-1937-4d7d-a123-6eeaa35ae5fd"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/8\n",
            "88/88 [==============================] - 17s 61ms/step - loss: 0.1854 - accuracy: 0.8231 - val_loss: 0.1219 - val_accuracy: 0.8331\n",
            "Epoch 2/8\n",
            "88/88 [==============================] - 2s 26ms/step - loss: 0.0967 - accuracy: 0.8593 - val_loss: 0.0746 - val_accuracy: 0.8925\n",
            "Epoch 3/8\n",
            "88/88 [==============================] - 2s 25ms/step - loss: 0.0636 - accuracy: 0.9079 - val_loss: 0.0513 - val_accuracy: 0.9260\n",
            "Epoch 4/8\n",
            "88/88 [==============================] - 2s 26ms/step - loss: 0.0447 - accuracy: 0.9366 - val_loss: 0.0395 - val_accuracy: 0.9454\n",
            "Epoch 5/8\n",
            "88/88 [==============================] - 2s 26ms/step - loss: 0.0345 - accuracy: 0.9519 - val_loss: 0.0352 - val_accuracy: 0.9502\n",
            "Epoch 6/8\n",
            "88/88 [==============================] - 2s 26ms/step - loss: 0.0277 - accuracy: 0.9609 - val_loss: 0.0325 - val_accuracy: 0.9549\n",
            "Epoch 7/8\n",
            "88/88 [==============================] - 2s 25ms/step - loss: 0.0238 - accuracy: 0.9657 - val_loss: 0.0309 - val_accuracy: 0.9569\n",
            "Epoch 8/8\n",
            "88/88 [==============================] - 2s 27ms/step - loss: 0.0209 - accuracy: 0.9698 - val_loss: 0.0310 - val_accuracy: 0.9561\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<keras.callbacks.History at 0x7f8abc10fb90>"
            ]
          },
          "metadata": {},
          "execution_count": 27
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print(\"\\n 테스트 정확도: %.4f\" % (model.evaluate(X_test, y_test)[1]))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "BWSgr2IAnubM",
        "outputId": "f6f0474f-e24e-44e8-f5e7-5c639ddca94b"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "88/88 [==============================] - 1s 8ms/step - loss: 0.0310 - accuracy: 0.9561\n",
            "\n",
            " 테스트 정확도: 0.9561\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "i=10 # 확인하고 싶은 테스트용 샘플의 인덱스.\n",
        "y_predicted = model.predict(np.array([X_test[i]])) # 입력한 테스트용 샘플에 대해서 예측 y를 리턴\n",
        "y_predicted = np.argmax(y_predicted, axis=-1) # 확률값 벡터를 정수 인코딩으로 변경.\n",
        "true = np.argmax(y_test[i], -1) # 원-핫 벡터를 정수 인코딩으로 변경.\n",
        "\n",
        "print(\"{:15}|{:5}|{}\".format(\"단어\", \"실제값\", \"예측값\"))\n",
        "print(35 * \"-\")\n",
        "\n",
        "for w, t, pred in zip(X_test[i], true, y_predicted[0]):\n",
        "    if w != 0: # PAD값은 제외함.\n",
        "        print(\"{:17}: {:7} {}\".format(index_to_word[w], index_to_ner[t].upper(), index_to_ner[pred].upper()))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "06GD-is2nvvE",
        "outputId": "cc32888b-d0e6-4c6a-ce1e-87a5fe55ac66"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "단어             |실제값  |예측값\n",
            "-----------------------------------\n",
            "sarah            : B-PER   B-PER\n",
            "brady            : I-PER   I-PER\n",
            ",                : O       O\n",
            "whose            : O       O\n",
            "republican       : B-MISC  B-MISC\n",
            "husband          : O       O\n",
            "was              : O       O\n",
            "OOV              : O       O\n",
            "OOV              : O       O\n",
            "in               : O       O\n",
            "an               : O       O\n",
            "OOV              : O       O\n",
            "attempt          : O       O\n",
            "on               : O       O\n",
            "president        : O       O\n",
            "ronald           : B-PER   B-PER\n",
            "reagan           : I-PER   I-PER\n",
            ",                : O       O\n",
            "took             : O       O\n",
            "centre           : O       O\n",
            "stage            : O       O\n",
            "at               : O       O\n",
            "the              : O       O\n",
            "democratic       : B-MISC  B-MISC\n",
            "national         : I-MISC  I-MISC\n",
            "convention       : I-MISC  I-MISC\n",
            "on               : O       O\n",
            "monday           : O       O\n",
            "night            : O       O\n",
            "to               : O       O\n",
            "OOV              : O       O\n",
            "president        : O       O\n",
            "bill             : B-PER   B-PER\n",
            "clinton          : I-PER   I-PER\n",
            "'s               : O       O\n",
            "gun              : O       O\n",
            "control          : O       O\n",
            "efforts          : O       O\n",
            ".                : O       O\n"
          ]
        }
      ]
    }
  ]
}