训练集链接
提取码：axpf
训练集（正常邮件）结构截图：
训练集里面正常邮件normal和垃圾邮件spam各有24封，利用这些数据训练出模型并对两份待分类邮件进行分类。
邮件长这样：
关于如何利用朴素贝叶斯进行分类，请参考：朴素贝叶斯（Naive Bayes）原理+编程实现拉普拉斯修正的朴素贝叶斯分类器
分类实现过程：
首先需要对每一封邮件进行切割处理，得到包含所有词语的列表，处理方法请参考：Python读取有空行的txt文件+将内容分割保存到列表中训练模型，利用贝叶斯公式计算出后验概率得到结果
完整代码：
# -*- coding: utf-8 -*-"""@Time ： /12/19 19:48@Author ：KI@File ：NaiveBayes_email.py@Motto：Hungry And Humble"""# 读取所有训练数据并按照空格分隔，保存在一个列表里返回def load_file(path):cab = []for i in range(1, 25):data = open(path % i)for line in data.readlines():cab.append(line.strip().split(','))cab_f = []for i in range(len(cab)):for j in range(len(cab[i])):if cab[i][j] != '':cab_f.append(cab[i][j].strip())cab_final = []for i in cab_f:for j in i.split(' '):cab_final.append(j)return cab_final# 朴素贝叶斯分类器def bayes(sample):path1 = 'Emails/Training/normal/%d.txt'path2 = 'Emails/Training/spam/%d.txt'normal_data = load_file(path1)spam_data = load_file(path2)# 计算p(x|C1)=p1与p(x|C2)=p2p1 = 1.0p2 = 1.0for i in range(len(sample)):x = 0.0for j in normal_data:if sample[i] == j:x = x + 1.0p1 = p1 * ((x + 1.0) / (len(normal_data) + 2.0)) # 拉普拉斯平滑for i in range(len(sample)):x = 0.0for j in spam_data:if sample[i] == j:x = x + 1.0p2 = p2 * ((x + 1.0) / (len(spam_data) + 2.0)) # 拉普拉斯平滑pc1 = len(normal_data) / (len(normal_data) + len(spam_data))pc2 = 1 - pc1if p1 * pc1 > p2 * pc2:return 'normal'else:return 'spam'# 测试def test(path):data = open(path)cab = []for line in data.readlines():cab.append(line.strip().split(','))cab_f = []for i in range(len(cab)):for j in range(len(cab[i])):if cab[i][j] != '':cab_f.append(cab[i][j].strip())cab_final = []for i in cab_f:for j in i.split(' '):cab_final.append(j)return bayes(cab_final)if __name__ == '__main__':print(test('Emails/test/normal.txt'))print(test('Emails/test/spam.txt'))sum1 = 0sum2 = 0# 再试试训练集for i in range(1, 25):if test('Emails/Training/normal/%d.txt' % i) == 'normal':sum1 = sum1 + 1for i in range(1, 25):if test('Emails/Training/spam/%d.txt' % i) == 'spam':sum2 = sum2 + 1print('normal分类正确率：', sum1 / 24)print('spam分类正确率：', sum2 / 24)
运行结果：
normal
spam
normal分类正确率： 0.9583333333333334
spam分类正确率： 1.0
可以看到，分类效果还不错！！
