{ "cells": [ { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "# 基于字典的情感分析\n", "\n", "以下内容来自**邓旭东HIT** https://zhuanlan.zhihu.com/p/23225934\n", "\n", "情感分析就是分析一句话说得是很主观还是客观描述，分析这句话表达的是积极的情绪还是消极的情绪。" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "## 原理\n", "比如这么一句话：\n", "\n", "> “这手机的画面极好，操作也比较流畅。不过拍照真的太烂了！系统也不好。”\n", "\n" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "① 情感词\n", "\n", "要分析一句话是积极的还是消极的，最简单最基础的方法就是找出句子里面的情感词，积极的情感词比如：赞，好，顺手，华丽等，消极情感词比如：差，烂，坏，坑爹等。出现一个积极词就+1，出现一个消极词就-1。\n", "里面就有“好”，“流畅”两个积极情感词，“烂”一个消极情感词。那它的情感分值就是1+1-1+1=2. 很明显这个分值是不合理的，下面一步步修改它。\n", "\n" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "subslide" } }, "source": [ "\n", "\n", "② 程度词\n", "\n", "“好”，“流畅”和‘烂“前面都有一个程度修饰词。”极好“就比”较好“或者”好“的情感更强，”太烂“也比”有点烂“情感强得多。所以需要在找到情感词后往前找一下有没有程度修饰，并给不同的程度一个权值。比如”极“，”无比“，”太“就要把情感分值*4，”较“，”还算“就情感分值*2，”只算“，”仅仅“这些就*0.5了。那么这句话的情感分值就是：4*1+1*2-1*4+1=3\n", "\n" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "subslide" } }, "source": [ "\n", "\n", "③ 感叹号\n", "\n", "可以发现太烂了后面有感叹号，叹号意味着情感强烈。因此发现叹号可以为情感值+2. 那么这句话的情感分值就变成了：4*1+1*2-1*4-2+1 = 1\n" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "subslide" } }, "source": [ "\n", "\n", "④ 否定词\n", "\n", "明眼人一眼就看出最后面那个”好“并不是表示”好“，因为前面还有一个”不“字。所以在找到情感词的时候，需要往前找否定词。比如”不“，”不能“这些词。而且还要数这些否定词出现的次数，如果是单数，情感分值就*-1，但如果是偶数，那情感就没有反转，还是*1。在这句话里面，可以看出”好“前面只有一个”不“，所以”好“的情感值应该反转，*-1。\n", "因此这句话的准确情感分值是：4*1+1*2-1*4-2+1*-1 = -1\n", "\n" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "subslide" } }, "source": [ "\n", "⑤ 积极和消极分开来\n", "\n", "再接下来，很明显就可以看出，这句话里面有褒有贬，不能用一个分值来表示它的情感倾向。而且这个权值的设置也会影响最终的情感分值，敏感度太高了。因此对这句话的最终的正确的处理，是得出这句话的一个积极分值，一个消极分值（这样消极分值也是正数，无需使用负数了）。它们同时代表了这句话的情感倾向。所以这句评论应该是”积极分值：6，消极分值：7“\n" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "subslide" } }, "source": [ "\n", "⑥ 以分句的情感为基础\n", "\n", "再仔细一步，详细一点，一条评论的情感分值是由不同的分句加起来的，因此要得到一条评论的情感分值，就要先计算出评论中每个句子的情感分值。这条例子评论有四个分句，因此其结构如下（[积极分值, 消极分值]）：[[4, 0], [2, 0], [0, 6], [0, 1]] \n", "\n", "以上就是使用情感词典来进行情感分析的主要流程了，算法的设计也会按照这个思路来实现。\n" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "### 算法设计\n", "- 第一步：读取评论数据，对评论进行分句。\n", "- 第二步：查找对分句的情感词，记录积极还是消极，以及位置。\n", "- 第三步：往情感词前查找程度词，找到就停止搜寻。为程度词设权值，乘以情感值。\n", "- 第四步：往情感词前查找否定词，找完全部否定词，若数量为奇数，乘以-1，若为偶数，乘以1。\n", "- 第五步：判断分句结尾是否有感叹号，有叹号则往前寻找情感词，有则相应的情感值+2。\n", "- 第六步：计算完一条评论所有分句的情感值，用数组（list）记录起来。\n", "- 第七步：计算并记录所有评论的情感值。\n", "- 第八步：通过分句计算每条评论的积极情感均值，消极情感均值，积极情感方差，消极情感方差。" ] }, { "cell_type": "code", "execution_count": 1, "metadata": { "ExecuteTime": { "end_time": "2019-06-14T03:38:44.934368Z", "start_time": "2019-06-14T03:38:44.788100Z" }, "slideshow": { "slide_type": "slide" } }, "outputs": [], "source": [ "import csv\n", "import jieba\n", "import numpy as np" ] }, { "cell_type": "code", "execution_count": 2, "metadata": { "ExecuteTime": { "end_time": "2019-06-14T03:38:50.400653Z", "start_time": "2019-06-14T03:38:50.386728Z" }, "slideshow": { "slide_type": "subslide" } }, "outputs": [], "source": [ "#打开词典文件，返回列表\n", "def open_dict(Dict, path):\n", " path = path + '%s.txt' % Dict\n", " dictionary = open(path, 'r', encoding='utf-8')\n", " dict = []\n", " for word in dictionary:\n", " word = word.strip('\\n')\n", " dict.append(word)\n", " return dict\n", "\n", "def judgeodd(num):\n", " if (num % 2) == 0:\n", " return 'even'\n", " else:\n", " return 'odd'" ] }, { "cell_type": "code", "execution_count": 4, "metadata": { "ExecuteTime": { "end_time": "2019-06-14T03:40:17.233566Z", "start_time": "2019-06-14T03:40:17.215170Z" }, "slideshow": { "slide_type": "subslide" } }, "outputs": [], "source": [ "#修改成自己的path路径。\n", "deny_word = open_dict(Dict = '否定词', path= r'../data/Textmining/')\n", "posdict = open_dict(Dict = 'positive', path= r'../data/Textmining/')\n", "negdict = open_dict(Dict = 'negative', path= r'../data/Textmining/')\n", "degree_word = open_dict(Dict = '程度级别词语', path= r'../data/Textmining/')" ] }, { "cell_type": "code", "execution_count": 5, "metadata": { "ExecuteTime": { "end_time": "2019-06-14T03:40:24.074618Z", "start_time": "2019-06-14T03:40:24.061928Z" } }, "outputs": [ { "data": { "text/plain": [ "(225, ['extreme', '百分之百', '倍加'])" ] }, "execution_count": 5, "metadata": {}, "output_type": "execute_result" } ], "source": [ "len(degree_word), degree_word[:3]" ] }, { "cell_type": "code", "execution_count": 6, "metadata": { "ExecuteTime": { "end_time": "2019-06-14T03:40:54.184650Z", "start_time": "2019-06-14T03:40:54.178860Z" }, "slideshow": { "slide_type": "subslide" } }, "outputs": [], "source": [ "mostdict = degree_word[degree_word.index('extreme')+1 : degree_word.index('very')]#权重4，即在情感词前乘以3\n", "verydict = degree_word[degree_word.index('very')+1 : degree_word.index('more')]#权重3\n", "moredict = degree_word[degree_word.index('more')+1 : degree_word.index('ish')]#权重2\n", "ishdict = degree_word[degree_word.index('ish')+1 : degree_word.index('last')]#权重0.5" ] }, { "cell_type": "code", "execution_count": 8, "metadata": { "ExecuteTime": { "end_time": "2019-06-14T03:41:06.028706Z", "start_time": "2019-06-14T03:41:06.023829Z" } }, "outputs": [ { "data": { "text/plain": [ "['百分之百', '倍加', '备至', '不得了', '不堪']" ] }, "execution_count": 8, "metadata": {}, "output_type": "execute_result" } ], "source": [ "mostdict[:5]" ] }, { "cell_type": "code", "execution_count": 10, "metadata": { "ExecuteTime": { "end_time": "2019-06-14T03:41:55.220611Z", "start_time": "2019-06-14T03:41:54.941723Z" }, "code_folding": [], "slideshow": { "slide_type": "subslide" } }, "outputs": [], "source": [ "def sentiment_score_list(dataset):\n", " seg_sentence = dataset.split('。')\n", " count1 = []\n", " count2 = []\n", " for sen in seg_sentence: #循环遍历每一个评论\n", " segtmp = jieba.lcut(sen, cut_all=False) #把句子进行分词，以列表的形式返回\n", " i = 0 #记录扫描到的词的位置\n", " a = 0 #记录情感词的位置\n", " poscount = 0 #积极词的第一次分值\n", " poscount2 = 0 #积极词反转后的分值\n", " poscount3 = 0 #积极词的最后分值（包括叹号的分值）\n", " negcount = 0\n", " negcount2 = 0\n", " negcount3 = 0\n", " for word in segtmp:\n", " if word in posdict: # 判断词语是否是情感词\n", " poscount += 1\n", " c = 0\n", " for w in segtmp[a:i]: # 扫描情感词前的程度词\n", " if w in mostdict:\n", " poscount *= 4.0\n", " elif w in verydict:\n", " poscount *= 3.0\n", " elif w in moredict:\n", " poscount *= 2.0\n", " elif w in ishdict:\n", " poscount *= 0.5\n", " elif w in deny_word:\n", " c += 1\n", " if judgeodd(c) == 'odd': # 扫描情感词前的否定词数\n", " poscount *= -1.0\n", " poscount2 += poscount\n", " poscount = 0\n", " poscount3 = poscount + poscount2 + poscount3\n", " poscount2 = 0\n", " else:\n", " poscount3 = poscount + poscount2 + poscount3\n", " poscount = 0\n", " a = i + 1 # 情感词的位置变化\n", "\n", " elif word in negdict: # 消极情感的分析，与上面一致\n", " negcount += 1\n", " d = 0\n", " for w in segtmp[a:i]:\n", " if w in mostdict:\n", " negcount *= 4.0\n", " elif w in verydict:\n", " negcount *= 3.0\n", " elif w in moredict:\n", " negcount *= 2.0\n", " elif w in ishdict:\n", " negcount *= 0.5\n", " elif w in degree_word:\n", " d += 1\n", " if judgeodd(d) == 'odd':\n", " negcount *= -1.0\n", " negcount2 += negcount\n", " negcount = 0\n", " negcount3 = negcount + negcount2 + negcount3\n", " negcount2 = 0\n", " else:\n", " negcount3 = negcount + negcount2 + negcount3\n", " negcount = 0\n", " a = i + 1\n", " elif word == '！' or word == '!': ##判断句子是否有感叹号\n", " for w2 in segtmp[::-1]: # 扫描感叹号前的情感词，发现后权值+2，然后退出循环\n", " if w2 in posdict or negdict:\n", " poscount3 += 2\n", " negcount3 += 2\n", " break\n", " i += 1 # 扫描词位置前移\n", "\n", "\n", " # 防止出现负数的情况\n", " pos_count = 0\n", " neg_count = 0\n", " if poscount3 < 0 and negcount3 > 0:\n", " neg_count += negcount3 - poscount3\n", " pos_count = 0\n", " elif negcount3 < 0 and poscount3 > 0:\n", " pos_count = poscount3 - negcount3\n", " neg_count = 0\n", " elif poscount3 < 0 and negcount3 < 0:\n", " neg_count = -poscount3\n", " pos_count = -negcount3\n", " else:\n", " pos_count = poscount3\n", " neg_count = negcount3\n", "\n", " count1.append([pos_count, neg_count])\n", " count2.append(count1)\n", " count1 = []\n", "\n", " return count2" ] }, { "cell_type": "code", "execution_count": 9, "metadata": { "ExecuteTime": { "end_time": "2019-06-14T03:41:50.740957Z", "start_time": "2019-06-14T03:41:50.733483Z" }, "code_folding": [], "slideshow": { "slide_type": "subslide" } }, "outputs": [], "source": [ "def sentiment_score(senti_score_list):\n", " score = []\n", " for review in senti_score_list:\n", " score_array = np.matrix(review)\n", " Pos = np.sum(score_array,axis=0)\n", " Neg = np.sum(score_array[:, 1])\n", " score.append([Pos, Neg])\n", " return score" ] }, { "cell_type": "code", "execution_count": 12, "metadata": { "ExecuteTime": { "end_time": "2019-06-14T03:42:48.432186Z", "start_time": "2019-06-14T03:42:48.413544Z" }, "slideshow": { "slide_type": "subslide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "[[matrix([[ 78., 169.]]), 169.0]]\n", "[[matrix([[498., 48.]]), 48.0]]\n" ] } ], "source": [ "data = '你就是个王八蛋，混账玩意!你们的手机真不好用！非常生气，我非常郁闷！！！！'\n", "data2= '我好开心啊，非常非常非常高兴！今天我得了一百分，我很兴奋开心，愉快，开心!!!'\n", "\n", "print(sentiment_score(sentiment_score_list(data)))\n", "print(sentiment_score(sentiment_score_list(data2)))" ] }, { "cell_type": "markdown", "metadata": { "slideshow": { "slide_type": "slide" } }, "source": [ "# 利用朴素贝叶斯算法进行情感分析\n" ] }, { "cell_type": "code", "execution_count": 13, "metadata": { "ExecuteTime": { "end_time": "2019-06-14T03:43:56.431397Z", "start_time": "2019-06-14T03:43:55.089588Z" }, "code_folding": [], "slideshow": { "slide_type": "subslide" } }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "I feel happy this morning ： neg\n", "Oh I love my friend ： pos\n", "not like that man ： neg\n", "this hourse is not great ： neg\n", "your song is annoying ： neg\n" ] } ], "source": [ "from textblob.classifiers import NaiveBayesClassifier\n", "\n", "train=[\n", " ('I love this car','pos'),\n", " ('This view is amazing','pos'),\n", " ('I feel great','pos'),\n", " ('I am so excited about the concert','pos'),\n", " (\"He is my best friend\",'pos'),\n", " ('I do not like this car','neg'),\n", " ('This view is horrible','neg'),\n", " (\"I feel tired this morning\",'neg'),\n", " ('I am not looking forward to the concert','neg'),\n", " ('He is an annoying enemy','neg')\n", "]\n", "\n", "test=[\n", " ('I feel happy this morning','pos'),\n", " ('Oh I love my friend','pos'),\n", " ('not like that man','neg'),\n", " (\"this hourse is not great\",'neg'),\n", " ('your song is annoying','neg')\n", "]\n", "\n", "cl=NaiveBayesClassifier(train)\n", "\n", "for sentence in test:\n", " print(sentence[0],'：',cl.classify(sentence[0]))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "celltoolbar": "Slideshow", "kernelspec": { "display_name": "Python [default]", "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": "3.5.4" }, "latex_envs": { "LaTeX_envs_menu_present": true, "autoclose": false, "autocomplete": true, "bibliofile": "biblio.bib", "cite_by": "apalike", "current_citInitial": 1, "eqLabelWithNumbers": true, "eqNumInitial": 1, "hotkeys": { "equation": "Ctrl-E", "itemize": "Ctrl-I" }, "labels_anchors": false, "latex_user_defs": false, "report_style_numbering": false, "user_envs_cfg": false }, "toc": { "base_numbering": 1, "nav_menu": {}, "number_sections": false, "sideBar": true, "skip_h1_title": false, "title_cell": "Table of Contents", "title_sidebar": "Contents", "toc_cell": false, "toc_position": {}, "toc_section_display": true, "toc_window_display": false } }, "nbformat": 4, "nbformat_minor": 2 }