機器學習之垃圾郵件分類2

讀取

def read_dataset(file_path='../data/SMSSpamCollection'):
    """
    讀取數據集
    :return: 返回數據和標題
    """

    with open(file_path, encoding='utf-8') as f:  # 讀取數據
        # 存儲標題
        sms_label = []
        # 存儲數據
        sms_data = []
        csv_reader = csv.reader(f, delimiter='\t')
        for line in csv_reader:
            sms_label.append(line[0])  # 提取出標籤
            sms_data.append(preprocessing(line[1]))  # 對每封郵件做預處理
    return sms_data, sms_label

數據預處理

def preprocessing(text):
    """
    預處理
    :param text:
    :return:
    """
    tokens = [word for sent in nltk.sent_tokenize(text) for word in nltk.word_tokenize(sent)]  # 分詞
    stops = stopwords.words('english')  # 使用英文的停用詞表
    tokens = [token for token in tokens if token not in stops]  # 停用詞
    tokens = [token.lower() for token in tokens if len(token) >= 3]  # 大小寫，短詞
    lmtzr = WordNetLemmatizer()
    tag = nltk.pos_tag(tokens)  # 詞性
    tokens = [lmtzr.lemmatize(token, pos=get_wordnet_pos(tag[i][1])) for i, token in enumerate(tokens)]  # 詞性還原
    preprocessed_text = ' '.join(tokens)
    return preprocessed_text

數據劃分—訓練集和測試集數據劃分

from sklearn.model_selection import train_test_split

x_train,x_test, y_train, y_test = train_test_split(data, target, test_size=0.2, random_state=0, stratify=y_train)

def split_dataset(data, label):
    """
    劃分訓練集和測試集
    :param data:
    :param label:
    :return:
    """
    x_train, x_test, y_train, y_test = train_test_split(data, label, test_size=0.2, random_state=0, stratify=label)
    return x_train, x_test, y_train, y_test

文本特徵提取

sklearn.feature_extraction.text.CountVectorizer

https://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.text.CountVectorizer.html?highlight=sklearn%20feature_extraction%20text%20tfidfvectorizer

sklearn.feature_extraction.text.TfidfVectorizer

https://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.text.TfidfVectorizer.html?highlight=sklearn%20feature_extraction%20text%20tfidfvectorizer#sklearn.feature_extraction.text.TfidfVectorizer

from sklearn.feature_extraction.text import TfidfVectorizer

tfidf2 = TfidfVectorizer()

觀察郵件與向量的關係

向量還原爲郵件

def tfidf_dataset(x_train, x_test):
    """
    把原始文本轉化爲tf-idf的特徵矩陣
    :param x_train:
    :param x_test:
    :return:
    """
    tfidf = TfidfVectorizer()
    X_train = tfidf.fit_transform(x_train)  # X_train用fit_transform生成詞彙表
    X_test = tfidf.transform(x_test)  # X_test要與X_train詞彙表相同，因此在X_train進行fit_transform基礎上進行transform操作
    return X_train, X_test, tfidf


def revert_mail(x_train, X_train, model):
    """
    向量還原郵件
    :param x_train:
    :param X_train:
    :param model:
    :return:
    """
    s = X_train.toarray()[0]
    print("第一封郵件向量表示爲：", s)
    # 該函數輸入一個矩陣，返回扁平化後矩陣中非零元素的位置（index）
    a = np.flatnonzero(X_train.toarray()[0])  # 非零元素的位置（index）
    print("非零元素的位置:", a)
    print("向量的非零元素的值：", s[a])
    b = model.vocabulary_  # 詞彙表
    key_list = []
    for key, value in b.items():
        if value in a:
            key_list.append(key)  # key非0元素對應的單詞
    print("向量非零元素對應的單詞：", key_list)
    print("向量化之前的郵件：", x_train[0])

 

模型選擇

from sklearn.naive_bayes import GaussianNB

from sklearn.naive_bayes import MultinomialNB

說明爲什麼選擇這個模型？

def mnb_model(x_train, x_test, y_train, y_test):
    """
    模型選擇（根據數據特點選擇多項式分佈）
    :param x_train:
    :param x_test:
    :param y_train:
    :param y_test:
    :return:
    """
    mnb = MultinomialNB()
    mnb.fit(x_train, y_train)
    ypre_mnb = mnb.predict(x_test)
    print("總數：", len(y_test))
    print("預測值正確數：", (ypre_mnb == y_test).sum())
    return ypre_mnb

模型評價：混淆矩陣，分類報告

from sklearn.metrics import confusion_matrix

confusion_matrix = confusion_matrix(y_test, y_predict)

說明混淆矩陣的含義

from sklearn.metrics import classification_report

說明準確率、精確率、召回率、F值分別代表的意義

def class_report(ypre_mnb, y_test):
    """
    模型評價：混淆矩陣，分類報告
    :param ypre_mnb:
    :param y_test:
    :return:
    """
    conf_matrix = confusion_matrix(y_test, ypre_mnb)
    print("混淆矩陣：\n", conf_matrix)
    c = classification_report(y_test, ypre_mnb)
    print("------------------------------------------")
    print("分類報告：\n", c)
    print("模型準確率：%.2f%%"%((conf_matrix[0][0] + conf_matrix[1][1]) / np.sum(conf_matrix)*100))

比較與總結

如果用CountVectorizer進行文本特徵生成，與TfidfVectorizer相比，效果如何？

 前者只考慮詞彙在文本中出現的頻率，屬於詞袋模型特徵，後者除了考量某詞彙在文本出現的頻率，還關注包含這個詞彙的所有文本的數量，能夠削減高頻沒有意義的詞彙出現帶來的影響, 挖掘更有意義的特徵。屬於Tfidf特徵。兩者相比，對於負類的預測更加準確，而正類的預測則稍遜色。但總體預測正確率也比TfidfVectorizer稍高，相比之下似乎CountVectorizer更適合進行預測。

完整代碼

"""
 @author Rakers
 @guide 郵件處理2
"""


import nltk, csv
import numpy as np
from nltk.corpus import stopwords
from nltk.stem import WordNetLemmatizer
from sklearn.model_selection import train_test_split
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.naive_bayes import MultinomialNB
from sklearn.metrics import confusion_matrix, classification_report


def get_wordnet_pos(treebank_tag):
    """
    根據詞性，生成還原參數pos
    :param treebank_tag:
    :return:
    """
    if treebank_tag.startswith('J'):  # adj
        return nltk.corpus.wordnet.ADJ
    elif treebank_tag.startswith('V'):  # v
        return nltk.corpus.wordnet.VERB
    elif treebank_tag.startswith('N'):  # n
        return nltk.corpus.wordnet.NOUN
    elif treebank_tag.startswith('R'):  # adv
        return nltk.corpus.wordnet.ADV
    else:
        return nltk.corpus.wordnet.NOUN


def preprocessing(text):
    """
    預處理
    :param text:
    :return:
    """
    tokens = [word for sent in nltk.sent_tokenize(text) for word in nltk.word_tokenize(sent)]  # 分詞
    stops = stopwords.words('english')  # 使用英文的停用詞表
    tokens = [token for token in tokens if token not in stops]  # 停用詞
    tokens = [token.lower() for token in tokens if len(token) >= 3]  # 大小寫，短詞
    lmtzr = WordNetLemmatizer()
    tag = nltk.pos_tag(tokens)  # 詞性
    tokens = [lmtzr.lemmatize(token, pos=get_wordnet_pos(tag[i][1])) for i, token in enumerate(tokens)]  # 詞性還原
    preprocessed_text = ' '.join(tokens)
    return preprocessed_text


def read_dataset():
    """
    讀取數據集
    :return: 返回數據和標題
    """
    file_path = r'SMSSpamCollection'
    sms = open(file_path, encoding='utf-8')  # 讀取數據
    # 存儲標題
    sms_label = []
    # 存儲數據
    sms_data = []
    csv_reader = csv.reader(sms, delimiter='\t')
    for line in csv_reader:
        sms_label.append(line[0])  # 提取出標籤
        sms_data.append(preprocessing(line[1]))  # 對每封郵件做預處理
    sms.close()
    return sms_data, sms_label


def split_dataset(data, label):
    """
    劃分訓練集和測試集
    :param data:
    :param label:
    :return:
    """
    x_train, x_test, y_train, y_test = train_test_split(data, label, test_size=0.2, random_state=0, stratify=label)
    return x_train, x_test, y_train, y_test

def tfidf_dataset(x_train, x_test):
    """
    把原始文本轉化爲tf-idf的特徵矩陣
    :param x_train:
    :param x_test:
    :return:
    """
    tfidf = TfidfVectorizer()
    X_train = tfidf.fit_transform(x_train)  # X_train用fit_transform生成詞彙表
    X_test = tfidf.transform(x_test)  # X_test要與X_train詞彙表相同，因此在X_train進行fit_transform基礎上進行transform操作
    return X_train, X_test, tfidf


def revert_mail(x_train, X_train, model):
    """
    向量還原郵件
    :param x_train:
    :param X_train:
    :param model:
    :return:
    """
    s = X_train.toarray()[0]
    print("第一封郵件向量表示爲：", s)
    # 該函數輸入一個矩陣，返回扁平化後矩陣中非零元素的位置（index）
    a = np.flatnonzero(X_train.toarray()[0])  # 非零元素的位置（index）
    print("非零元素的位置:", a)
    print("向量的非零元素的值：", s[a])
    b = model.vocabulary_  # 詞彙表
    key_list = []
    for key, value in b.items():
        if value in a:
            key_list.append(key)  # key非0元素對應的單詞
    print("向量非零元素對應的單詞：", key_list)
    print("向量化之前的郵件：", x_train[0])


def mnb_model(x_train, x_test, y_train, y_test):
    """
    模型選擇（根據數據特點選擇多項式分佈）
    :param x_train:
    :param x_test:
    :param y_train:
    :param y_test:
    :return:
    """
    mnb = MultinomialNB()
    mnb.fit(x_train, y_train)
    ypre_mnb = mnb.predict(x_test)
    print("總數：", len(y_test))
    print("預測值正確數：", (ypre_mnb == y_test).sum())
    return ypre_mnb

def class_report(ypre_mnb, y_test):
    """
    模型評價：混淆矩陣，分類報告
    :param ypre_mnb:
    :param y_test:
    :return:
    """
    conf_matrix = confusion_matrix(y_test, ypre_mnb)
    print("混淆矩陣：\n", conf_matrix)
    c = classification_report(y_test, ypre_mnb)
    print("------------------------------------------")
    print("分類報告：\n", c)
    print("模型準確率：%.2f%%"%((conf_matrix[0][0] + conf_matrix[1][1]) / np.sum(conf_matrix)*100))


if __name__ == '__main__':
    # 讀取數據集
    sms_data, sms_label = read_dataset()
    # 劃分數據集
    x_train, x_test, y_train, y_test = split_dataset(sms_data, sms_label)
    # 把原始文本轉化爲tf-idf的特徵矩陣
    X_train, X_test, tfidf = tfidf_dataset(x_train, x_test)
    # 向量還原成郵件
    revert_mail(x_train, X_train, tfidf)
    # 模型選擇
    y_mnb = mnb_model(X_train, X_test, y_train, y_test)
    # 模型評價
    class_report(y_mnb, y_test)