{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# 朴素贝叶斯模型实验 （选作）\n", "\n", "> 本实验目标是用朴素贝叶斯模型对Yelp网站的评论文本进行分类" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 第一步：读入数据\n", "\n", "把`yelp.csv`读入一个DataFrame中。" ] }, { "cell_type": "code", "execution_count": 66, "metadata": {}, "outputs": [ { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
business_iddatereview_idstarstexttypeuser_idcoolusefulfunny
09yKzy9PApeiPPOUJEtnvkg2011-01-26fWKvX83p0-ka4JS3dc6E5A5My wife took me here on my birthday for breakf...reviewrLtl8ZkDX5vH5nAx9C3q5Q250
1ZRJwVLyzEJq1VAihDhYiow2011-07-27IjZ33sJrzXqU-0X6U8NwyA5I have no idea why some people give bad review...review0a2KyEL0d3Yb1V6aivbIuQ000
26oRAC4uyJCsJl1X0WZpVSA2012-06-14IESLBzqUCLdSzSqm0eCSxQ4love the gyro plate. Rice is so good and I als...review0hT2KtfLiobPvh6cDC8JQg010
3_1QQZuf4zZOyFCvXc0o6Vg2010-05-27G-WvGaISbqqaMHlNnByodA5Rosie, Dakota, and I LOVE Chaparral Dog Park!!...reviewuZetl9T0NcROGOyFfughhg120
46ozycU1RpktNG2-1BroVtw2012-01-051uJFq2r5QfJG_6ExMRCaGw5General Manager Scott Petello is a good egg!!!...reviewvYmM4KTsC8ZfQBg-j5MWkw000
\n", "
" ], "text/plain": [ " business_id date review_id stars \\\n", "0 9yKzy9PApeiPPOUJEtnvkg 2011-01-26 fWKvX83p0-ka4JS3dc6E5A 5 \n", "1 ZRJwVLyzEJq1VAihDhYiow 2011-07-27 IjZ33sJrzXqU-0X6U8NwyA 5 \n", "2 6oRAC4uyJCsJl1X0WZpVSA 2012-06-14 IESLBzqUCLdSzSqm0eCSxQ 4 \n", "3 _1QQZuf4zZOyFCvXc0o6Vg 2010-05-27 G-WvGaISbqqaMHlNnByodA 5 \n", "4 6ozycU1RpktNG2-1BroVtw 2012-01-05 1uJFq2r5QfJG_6ExMRCaGw 5 \n", "\n", " text type \\\n", "0 My wife took me here on my birthday for breakf... review \n", "1 I have no idea why some people give bad review... review \n", "2 love the gyro plate. Rice is so good and I als... review \n", "3 Rosie, Dakota, and I LOVE Chaparral Dog Park!!... review \n", "4 General Manager Scott Petello is a good egg!!!... review \n", "\n", " user_id cool useful funny \n", "0 rLtl8ZkDX5vH5nAx9C3q5Q 2 5 0 \n", "1 0a2KyEL0d3Yb1V6aivbIuQ 0 0 0 \n", "2 0hT2KtfLiobPvh6cDC8JQg 0 1 0 \n", "3 uZetl9T0NcROGOyFfughhg 1 2 0 \n", "4 vYmM4KTsC8ZfQBg-j5MWkw 0 0 0 " ] }, "execution_count": 66, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# read csv\n", "import pandas as pd\n", "\n", "url = \"e2.4_yelp.csv\"\n", "yelp = pd.read_csv(url)\n", "yelp.head()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "创建一个新的DataFrame，只包含5颗星和1颗星评分的数据。" ] }, { "cell_type": "code", "execution_count": 67, "metadata": {}, "outputs": [], "source": [ "# filter data\n", "yelp_best_worst = yelp[(yelp.stars==5) | (yelp.stars==1)]\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 第二步：生成X和y\n", "\n", "使用评论文本作为唯一的分类特征，评分星数作为预测目标，并将数据集划分为训练集和测试集。" ] }, { "cell_type": "code", "execution_count": 68, "metadata": {}, "outputs": [], "source": [ "# define X and y \n", "X = yelp_best_worst.text\n", "y = yelp_best_worst.stars\n" ] }, { "cell_type": "code", "execution_count": 69, "metadata": {}, "outputs": [], "source": [ "# split into training and testing sets\n", "from sklearn.model_selection import train_test_split\n", "X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=1)\n", "\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 第三步：转换数据\n", "\n", "使用CountVectorizer将X_train和X_test转换为document-term矩阵。" ] }, { "cell_type": "code", "execution_count": 70, "metadata": {}, "outputs": [], "source": [ "# import and instantiate the vectorizer\n", "from sklearn.feature_extraction.text import CountVectorizer\n", "vect = CountVectorizer()" ] }, { "cell_type": "code", "execution_count": 71, "metadata": {}, "outputs": [], "source": [ "# fit and transform X_train, but only transform X_test\n", "X_train_dtm = vect.fit_transform(X_train)\n", "X_test_dtm = vect.transform(X_test)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 第四步：训练、预测、评价\n", "\n", "使用朴素贝叶斯预测测试集中评论的星级评分，并计算预测精度。" ] }, { "cell_type": "code", "execution_count": 72, "metadata": {}, "outputs": [ { "data": { "text/html": [ "
MultinomialNB()
In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
" ], "text/plain": [ "MultinomialNB()" ] }, "execution_count": 72, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# import/instantiate/fit\n", "from sklearn.naive_bayes import MultinomialNB\n", "nb = MultinomialNB()\n", "nb.fit(X_train_dtm, y_train)" ] }, { "cell_type": "code", "execution_count": 73, "metadata": {}, "outputs": [], "source": [ "# make class predictions\n", "y_pred_class = nb.predict(X_test_dtm)" ] }, { "cell_type": "code", "execution_count": 74, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "0.9187866927592955" ] }, "execution_count": 74, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# calculate accuracy\n", "from sklearn import metrics\n", "metrics.accuracy_score(y_test, y_pred_class)\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "计算AUC。注意：y_test中的取值是1和5，需要先把它转换为取值为0和1的二值数组y_test_binary。" ] }, { "cell_type": "code", "execution_count": 75, "metadata": {}, "outputs": [], "source": [ "# create y_test_binary from y_test, which contains ones and zeros instead\n", "# of ones and fives\n", "y_test_binary = y_test.map({5:1, 1:0})\n" ] }, { "cell_type": "code", "execution_count": 76, "metadata": {}, "outputs": [], "source": [ "# predict class probabilities\n", "y_pred_prob = nb.predict_proba(X_test_dtm)[:, 1]" ] }, { "cell_type": "code", "execution_count": 77, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "0.9391635104285566" ] }, "execution_count": 77, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# calculate the AUC using y_test_binary and y_pred_prob\n", "metrics.roc_auc_score(y_test_binary, y_pred_prob)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "绘制ROC曲线。" ] }, { "cell_type": "code", "execution_count": 78, "metadata": {}, "outputs": [], "source": [ "%matplotlib inline\n", "import matplotlib.pyplot as plt" ] }, { "cell_type": "code", "execution_count": 79, "metadata": {}, "outputs": [ { "data": { "image/png": "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", "text/plain": [ "
" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "# plot ROC curve using y_test_binary and y_pred_prob\n", "fpr, tpr, thresholds = metrics.roc_curve(y_test_binary, y_pred_prob)\n", "plt.plot(fpr, tpr)\n", "plt.xlim([0.0, 1.0])\n", "plt.ylim([0.0, 1.0])\n", "plt.title('ROC curve for Yelp classifier')\n", "plt.xlabel('False Positive Rate (1 - Specificity)')\n", "plt.ylabel('True Positive Rate (Sensitivity)')\n", "plt.grid(True)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "显示混淆矩阵，并计算敏感度和特异性，评论结果。" ] }, { "cell_type": "code", "execution_count": 80, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "array([[126, 58],\n", " [ 25, 813]])" ] }, "execution_count": 80, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# print the confusion matrix\n", "metrics.confusion_matrix(y_test, y_pred_class)\n" ] }, { "cell_type": "code", "execution_count": 81, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "0.9701670644391408" ] }, "execution_count": 81, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# calculate sensitivity\n", "tn, fp, fn, tp = metrics.confusion_matrix(y_test, y_pred_class).ravel()\n", "sensitivity = tp / float(tp + fn)\n", "sensitivity" ] }, { "cell_type": "code", "execution_count": 82, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "0.6847826086956522" ] }, "execution_count": 82, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# calculate specificity\n", "specificity = tn / float(tn + fp)\n", "specificity\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "对模型的敏感度和特异性做出评论：" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 第五步：错误分析\n", "\n", "查看测试集中一些被预测错误的评论文本，即false positives和false negatives。试着回答为什么这些评论会被预测错。" ] }, { "cell_type": "code", "execution_count": 83, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "2175 This has to be the worst restaurant in terms o...\n", "1781 If you like the stuck up Scottsdale vibe this ...\n", "2674 I'm sorry to be what seems to be the lone one ...\n", "9984 Went last night to Whore Foods to get basics t...\n", "3392 I found Lisa G's while driving through phoenix...\n", "8283 Don't know where I should start. Grand opening...\n", "2765 Went last week, and ordered a dozen variety. I...\n", "2839 Never Again,\\nI brought my Mountain Bike in (w...\n", "321 My wife and I live around the corner, hadn't e...\n", "1919 D-scust-ing.\n", "2490 Lazy Q CLOSED in 2010. New Owners cleaned up ...\n", "9125 La Grande Orange Grocery has a problem. It can...\n", "9185 For frozen yogurt quality, I give this place a...\n", "436 this another place that i would give no stars ...\n", "2051 Sadly with new owners comes changes on menu. ...\n", "1721 This is the closest to a New York hipster styl...\n", "3447 If you want a school that cares more about you...\n", "842 Boy is the name a temptation.Seriously :) I'l...\n", "6159 Really, if I could, I would give this place ze...\n", "943 Don't waste your time...Arrowhead mall on the ...\n", "5977 You want good food? You'd be better off smuggl...\n", "8833 The owner has changed hands & this place isn't...\n", "6584 Jimmy Johns is cheaper and better ... The Capr...\n", "1899 Buca Di Beppo is literally, italian restaurant...\n", "9953 \"Hipster,Trendy\" ????-I think NOT !!!! Very di...\n", "2060 This place is closed. Good riddance.\n", "3082 Currently having a liquidation sale, but it's ...\n", "8220 Maybe I ate at a different restaurant than the...\n", "3634 Seriously?! With grocery stores like Fresh & E...\n", "3266 Absolutely awful... these guys have NO idea wh...\n", "7397 This place sucks!! I moved to the valley and h...\n", "4473 It is what you would expect from any themed pl...\n", "5502 Angry Bro Bar ! Please go here if you wear si...\n", "2615 Great in its day, now leaves a lot to be desir...\n", "3413 I purchased the Enotria groupon when it was re...\n", "2999 I can't even believe I actually went to this r...\n", "1372 No offense to everyone who gave this place 5 s...\n", "1291 Every time I come here the staff is so rude! I...\n", "6222 My mother always told me, if I didn't have any...\n", "9296 My boyfriend and I tried this place last year ...\n", "7975 What are you all talking about?! This place is...\n", "4630 I used to always go here for tires until my me...\n", "7130 I was not impressed. The food was bad & expens...\n", "5818 Most horrible buffet I have ever been to.\\n\\nM...\n", "3704 Staff is nice, and that's it.\\nThey use very c...\n", "8741 They served us stale rice. Average main dishe...\n", "3938 We were so disappointed! We were on vacation ...\n", "7631 this is a business located in the fry's grocer...\n", "8681 As I promised myself, I'd go back again to try...\n", "1532 Cold, under done chips. If a Mexican food rest...\n", "113 Unless you are a regular or look like your wal...\n", "4165 OMG! what is the rave about? this place is dis...\n", "9299 The salad plates were not chilled... As they u...\n", "4311 Donuts are really good, if they have any when ...\n", "7035 Totally excited to try this place out, my gran...\n", "8000 Still a place that is unacceptable in my book-...\n", "3755 Have been going to LGO since 2003 and have alw...\n", "507 HELLISH HELLISH SUMMER WEATHER (March thru Oct...\n", "Name: text, dtype: object" ] }, "execution_count": 83, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# first 10 false positives (meaning they were incorrectly classified as 5-star reviews)\n", "X_test[y_pred_class > y_test]\n" ] }, { "cell_type": "code", "execution_count": 84, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "7148 I now consider myself an Arizonian. If you dri...\n", "4963 This is by far my favourite department store, ...\n", "6318 Since I have ranted recently on poor customer ...\n", "380 This is a must try for any Mani Pedi fan. I us...\n", "5565 I`ve had work done by this shop a few times th...\n", "3448 I was there last week with my sisters and whil...\n", "6050 I went to sears today to check on a layaway th...\n", "2504 I've passed by prestige nails in walmart 100s ...\n", "2475 This place is so great! I am a nanny and had t...\n", "241 I was sad to come back to lai lai's and they n...\n", "3149 I was told to see Greg after a local shop diag...\n", "423 These guys helped me out with my rear windshie...\n", "763 Here's the deal. I said I was done with OT, bu...\n", "8956 I took my computer to RedSeven recently when m...\n", "750 This store has the most pleasant employees of ...\n", "9765 You can't give anything less than 5 stars to a...\n", "6334 I came here today for a manicure and pedicure....\n", "1282 Loved my haircut. Walked in and waited for jus...\n", "1266 I've never been to this location before. My hu...\n", "402 Once again Wildflower proves why it's my favor...\n", "4034 \"Fine dining\" is not just a setting. it isn't...\n", "2444 EXCELLENT CUSTOMER SERVICE! \\n\\nEven with Happ...\n", "2494 What a great surprise stumbling across this ba...\n", "5736 Thank goodness for Sue at Mill Avenue Travel. ...\n", "7903 First, I'm sorry this review is lengthy, but i...\n", "Name: text, dtype: object" ] }, "execution_count": 84, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# first 10 false negatives (meaning they were incorrectly classified as 1-star reviews)\n", "X_test[y_pred_class < y_test]" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 第六步：多分类预测\n", "\n", "使用所有的评论做预测，而不仅仅是评分1星和5星的评论。" ] }, { "cell_type": "code", "execution_count": 85, "metadata": {}, "outputs": [], "source": [ "# define X and y using the original DataFrame\n", "X = yelp.text\n", "y = yelp.stars\n" ] }, { "cell_type": "code", "execution_count": 86, "metadata": {}, "outputs": [], "source": [ "# split into training and testing sets\n", "X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=1)\n", "\n" ] }, { "cell_type": "code", "execution_count": 87, "metadata": {}, "outputs": [], "source": [ "# create document-term matrices\n", "X_train_dtm = vect.fit_transform(X_train)\n", "X_test_dtm = vect.transform(X_test)\n", "\n" ] }, { "cell_type": "code", "execution_count": 88, "metadata": {}, "outputs": [ { "data": { "text/html": [ "
MultinomialNB()
In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
" ], "text/plain": [ "MultinomialNB()" ] }, "execution_count": 88, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# fit a Naive Bayes model\n", "nb.fit(X_train_dtm, y_train)\n", "\n" ] }, { "cell_type": "code", "execution_count": 89, "metadata": {}, "outputs": [], "source": [ "# make class predictions\n", "y_pred_class = nb.predict(X_test_dtm)\n" ] }, { "cell_type": "code", "execution_count": 90, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "0.4712" ] }, "execution_count": 90, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# calculate the testing accuary\n", "metrics.accuracy_score(y_test, y_pred_class)\n" ] }, { "cell_type": "code", "execution_count": 91, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "array([[ 55, 14, 24, 65, 27],\n", " [ 28, 16, 41, 122, 27],\n", " [ 5, 7, 35, 281, 37],\n", " [ 7, 0, 16, 629, 232],\n", " [ 6, 4, 6, 373, 443]])" ] }, "execution_count": 91, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# print the confusion matrix\n", "metrics.confusion_matrix(y_test, y_pred_class)\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "有何评论：" ] } ], "metadata": { "kernelspec": { "display_name": "Python 3.10.8 ('.venv': venv)", "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": "3.10.8" }, "vscode": { "interpreter": { "hash": "1f0d395e06aa83586067b19165efc9b683889967164248deef4bbf1fa27cfb00" } } }, "nbformat": 4, "nbformat_minor": 1 }