{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<br></br>\n",
    "# **gensim | doc2vec**\n",
    "<br></br>\n",
    "## **1 네이버 리뷰 문장의 활용**\n",
    "네이버 영화리뷰 단어모델 만들기"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from konlpy.tag import Twitter\n",
    "twitter = Twitter()\n",
    "\n",
    "# 네이버 영화리뷰 데이터 불러오기 (1/10만 추출한다)\n",
    "def read_data(filename):\n",
    "    with open(filename, 'r', encoding='utf-8') as f:\n",
    "        data = [line.split('\\t') for line in f.read().splitlines()]\n",
    "        \n",
    "    from random import randint\n",
    "    random_data = [data[randint(1, len(data))]  for no in range(int(len(data)/10)) ]\n",
    "    return random_data\n",
    "\n",
    "# 한글 Token에 품사정보를 덧붙이기\n",
    "def tokenize(doc):\n",
    "    return ['/'.join(t) for t in twitter.pos(doc, norm=True, stem=True)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "from collections import namedtuple\n",
    "train_data        = read_data('./data/ratings_train.txt')\n",
    "train_docs        = [(tokenize(row[1]), row[2]) for row in train_data[1:]]\n",
    "TaggedDocument    = namedtuple('TaggedDocument', 'words tags')\n",
    "tagged_train_docs = [TaggedDocument(d, [c]) for d, c in train_docs]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from pprint import pprint\n",
    "pprint(tagged_train_docs[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<br></br>\n",
    "## **2 doc2vec 파라미터 설정 및 학습**\n",
    "**[세부적인 파라미터 설정](http://hero4earth.com/blog/projects/2018/01/21/naver_movie_review/)**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "from gensim.models import doc2vec\n",
    "# doc_vectorizer = doc2vec.Doc2Vec(\n",
    "#     dm          = 0,     # PV-DBOW / default 1\n",
    "#     dbow_words  = 1,     # w2v simultaneous with DBOW d2v / default 0\n",
    "#     window      = 8,     # distance between the predicted word and context words\n",
    "#     vector_size = 300,   # vector size\n",
    "#     alpha       = 0.025, # learning-rate\n",
    "#     seed        = 1234,\n",
    "#     min_count   = 20,    # ignore with freq lower\n",
    "#     min_alpha   = 0.025, # min learning-rate\n",
    "#     workers     = 4,     # multi cpu\n",
    "#     hs          = 1,     # hierarchical softmax / default 0\n",
    "#     negative    = 10,    # negative sampling / default 5\n",
    "# )\n",
    "\n",
    "# Doc2Vec 모델을 정의\n",
    "doc_vectorizer = doc2vec.Doc2Vec(vector_size=300, alpha=0.025, min_alpha=0.025, seed=1234)\n",
    "\n",
    "# 정의된 모델에 Token 자료를 연결\n",
    "doc_vectorizer.build_vocab(tagged_train_docs)\n",
    "\n",
    "for epoch in range(10):\n",
    "    doc_vectorizer.train(tagged_train_docs, \n",
    "                         total_examples = doc_vectorizer.corpus_count, \n",
    "                         epochs = doc_vectorizer.epochs)\n",
    "    doc_vectorizer.alpha -= 0.002\n",
    "    doc_vectorizer.min_alpha = doc_vectorizer.alpha \n",
    "\n",
    "# 학습이 완료된 모델의 데이터를 저장한다\n",
    "doc_vectorizer.save('./data/doc2vec.model')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<br></br>\n",
    "## **3 doc2Vec 모델활용**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%reset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%who"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from gensim.models import doc2vec\n",
    "from pprint import pprint\n",
    "doc_vectorizer = doc2vec.Doc2Vec.load('./data/doc2vec.model')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "pprint(doc_vectorizer.wv.most_similar('공포/Noun'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "doc_vectorizer.wv.similarity('공포/Noun', 'ㅋㅋ/KoreanParticle')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "pprint(doc_vectorizer.wv.most_similar(positive=['여자/Noun', '공포/Noun'], \n",
    "                                      negative=['남자/Noun']))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "doc_vectorizer.infer_vector(['픽사/Noun', '최고/Noun', '명작/Noun'])[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "doc_vectorizer.infer_vector(['픽사/Noun', '최고/Noun', '명작/Noun']).sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "doc_vectorizer.infer_vector(['호러/Noun', '여자/Noun', '공포/Noun'])[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "doc_vectorizer.infer_vector(['호러/Noun', '여자/Noun', '공포/Noun']).sum()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}