朴素贝叶斯进行新闻分类

  1 import os
  2 import pandas as pd
  3 from sklearn.feature_extraction.text import TfidfVectorizer
  4 from sklearn.model_selection import train_test_split
  5 from sklearn.naive_bayes import MultinomialNB
  6 from sklearn.externals import joblib
  7 
  8 from two_naive_bayes.utils import cut_word, rep_invalid_char
  9 
 10 
 11 class NaiveBayes(object):
 12     """朴素贝叶斯进行新闻分类，现阶段只支持(科技、娱乐、时尚)三个类别的数据分类"""
 13     def __init__(self, title, content):
 14         print(‘__init__‘)
 15         self.excel_reader = os.path.join(os.path.join(os.path.pardir, ‘one_spider‘), ‘tx.xlsx‘)
 16         self.data = cut_word([rep_invalid_char(str(title)+str(content))])
 17 
 18     def read(self):
 19         """读取数据，返回特征值与目标值的列表"""
 20         print(‘read.....‘)
 21         df = pd.read_excel(self.excel_reader, ‘Sheet‘)
 22         title_list = df[‘title‘].tolist()
 23         content_list = df[‘content‘].tolist()
 24         target = df[‘class‘].tolist()
 25         data = [(str(title_list[i])+str(content_list[i])) for i in range(len(title_list))]
 26         data = cut_word(data)
 27         return data, target
 28 
 29     def data_split(self, data, target):
 30         """
 31         将数据分割成数据集与测试集
 32         :param data: 特征值
 33         :param target: 目标值
 34         :return: 数据集的特征值，测试集的特征值，数据集的目标值，测试集的目标值
 35         """
 36         print(‘data_split.....‘)
 37         x_train, x_test, y_train, y_test = train_test_split(data, target)
 38         return x_train, x_test, y_train, y_test
 39 
 40     def feature_extraction(self, x_train):
 41         """对训练集进行特征抽取，返回TfIdf处理后的数据和tf"""
 42         print(‘feature_extraction.....‘)
 43         tf = TfidfVectorizer()
 44         x_train = tf.fit_transform(x_train)
 45         self.tf = tf  # 保存此tf，后面进行特征抽取都要用到
 46         joblib.dump(tf, ‘tf_model.m‘)
 47         return x_train
 48 
 49     def bayes(self, x_train, y_train):
 50         """贝叶斯建模"""
 51         print(‘bayes.....‘)
 52         mlt = MultinomialNB()
 53         mlt.fit(x_train, y_train)
 54         return mlt
 55 
 56     def test(self, mlt, x_test, y_test):
 57         """
 58         预测，返回准确率
 59         :param mlt: 朴素贝叶斯训练的模型
 60         :param x_test: 测试集的原始数据
 61         :param y_test: 测试集的目标值
 62         :return: 模型的准确率，每次运行的结果是不一样的
 63         """
 64         print(‘test.....‘)
 65         x_test = self.tf.transform(x_test)  # 这里用fit_transform数据集的tf，保证维度等的统一
 66         y_predict = mlt.predict(x_test)
 67         same_num = 0
 68         for i in range(len(y_test)):
 69             if y_predict[i] == y_test[i]:
 70                 same_num += 1
 71         return float(same_num/len(y_test))
 72 
 73     def predict(self):
 74         """根据传入的文章标题和内容，预测该文章属于哪一个类别"""
 75         print(‘predict.....‘)
 76         # 如果本地模型存在，则从本地模型读取，否则重新进入训练过程，训练并保存模型，得出准确率，并输出文章类别
 77         condition_one = os.path.exists(os.path.join(os.path.curdir, ‘train_model.m‘))
 78         condition_two = os.path.exists(os.path.join(os.path.curdir, ‘tf_model.m‘))
 79         if all([condition_one, condition_two]) :
 80             model = joblib.load(‘train_model.m‘)
 81             self.tf = joblib.load(‘tf_model.m‘)
 82         else:
 83             data, target = self.read()
 84             x_train, x_test, y_train, y_test = self.data_split(data, target)
 85             x_train = self.feature_extraction(x_train)
 86             model = self.bayes(x_train, y_train)
 87             joblib.dump(model, ‘train_model.m‘)  # 保存训练出来的模型
 88             number = self.test(model, x_test, y_test)
 89             print(‘模型预测的准确率为：‘, number)
 90 
 91         data = self.tf.transform(self.data)
 92         predict = model.predict(data)
 93         print(‘预测该文章的类别为：‘, predict[0])
 94 
 95     def run(self):
 96         """
 97         run方法用来训练模型并保存至本地，和评估准确率，predict方法虽然也有这个功能，
 98         但是为了训练处一个正确率较大的模型，就需要用run方法反复训练比较后得出最佳的
 99         模型保存至本地，在之后的预测中就不用重新训练模型而浪费时间了
100         训练对比模型时，可传入空的title和content即可
101         """
102         # todo: =======================================================
103         # 1.读取爬虫爬取的excel表数据
104         # 2.使用jieba分词进行中文分词，得到总的特征值与目标值
105         # 3.进行数据分割，分为数据集与测试集
106         # 4.对数据集进行特征抽取 TfIdf，测试集作为测试数据也需要特征抽取
107         # 5.以训练集中的列表进行重要性统计
108         # 6.朴素贝叶斯算法训练模型
109         # 7.模型对测试集进行预测，得出准确率
110         print(‘run.....‘)
111         data, target = self.read()
112         x_train, x_test, y_train, y_test = self.data_split(data, target)
113         x_train = self.feature_extraction(x_train)
114         model = self.bayes(x_train, y_train)
115         joblib.dump(model, ‘train_model.m‘)  # 保存训练出来的模型
116         number = self.test(model, x_test, y_test)
117         print(‘模型预测的准确率为：‘, number)
118 
119 
120 
121 if __name__ == ‘__main__‘:
122     # 需要预测的文章标题和文章内容(可更改)
123     title = ‘蔚来与Mobileye结盟，合作产品将于2022年面世‘
124     with open(‘110.txt‘, ‘r‘) as f:
125         content = f.read()
126     # title: 文章标题， content: 文章正文内容
127     nb = NaiveBayes(title, content)
128     # nb.run()
129     nb.predict()

 1 import re
 2 import jieba
 3 
 4 
 5 def cut_word(data:list):
 6     """
 7     进行中文分词
 8     :param data: 待分词的数据，列表类型
 9     :return: 分词后的数据，列表类型
10     """
11     res = []
12     for s in data:
13         word_cut = ‘ ‘.join(list(jieba.cut(s)))
14         res.append(word_cut)
15     return res
16 
17 def rep_invalid_char(old:str):
18     """只保留字符串中的大小写字母，数字，汉字"""
19     invalid_char_re = r"[^0-9A-Za-z\u4e00-\u9fa5]"
20     return re.sub(invalid_char_re, "", old)