实现本文的文本数据可以在THUCTC下载也可以自己手动爬虫生成，
本文主要参考：https://blog.csdn.net/hao5335156/article/details/82716923
nb表示朴素贝叶斯
rf表示随机森林
lg表示逻辑回归
初学者（我）通过本程序的学习可以巩固python基础，学会python文本的处理，和分类器的调用。方便接下来的机器学习的学习。
各个参数直观的含义：
# -*- coding: utf-8 -*-
"""
Created on Thu Nov 29 13:00:46 2018
@author: caoqu
"""
import matplotlib.pyplot as plt
import random
import os
import jieba
from sklearn.naive_bayes import MultinomialNB as NB
from sklearn.linear_model.logistic import LogisticRegression as LR
from sklearn.ensemble import RandomForestClassifier as RF
# 文本处理 --> 生成训练集 测试集 词频集
def text_processor(text_path, test_size=0.2):
folder_list = os.listdir(text_path)
data_list=[] # 每个元素均为一篇文章
class_list=[] # 对应于每篇文章的类别
# 一个循环读取一个类别的文件夹
for folder in folder_list:
new_folder_path = os.path.join(text_path, folder) # 类别列表
# 由于THUCTC文本巨多，所以我从每个类别的文本列表中随机抽取200个文本用于训练和测试，可以自行修改
files = random.sample(os.listdir(new_folder_path), 200)
# 一个循环读取一篇文章
for file in files:
with open(os.path.join(new_folder_path, file), 'r', encoding='UTF-8') as fp:
raw = fp.read()
word_cut = jieba.cut(raw, cut_all=False) #精确模式切分文章
word_list = list(word_cut) # 一篇文章一个 word_list
data_list.append(word_list)
class_list.append(folder.encode('utf-8'))
# 划分训练集和测试集
# data_class_list[[word_list_one[], 体育], [word_list_two[], 财经], ..., [...]]
data_class_list = list(zip(data_list, class_list))
random.shuffle(data_class_list) # 打乱顺序
index = int(len(data_class_list) * test_size) + 1 # 训测比为 8:2
train_list = data_class_list[index:]
test_list = data_class_list[:index]
train_data_list, train_class_list = zip(*train_list) # (word_list_one[],...), (体育,...)
test_data_list, test_class_list = zip(*test_list)
# 统计词频 all_words_dict{"key_word_one":100, "key_word_two":200, ...}
all_words_dict = {}
for word_list in train_data_list:
for word in word_list:
if all_words_dict.get(word) != None:
all_words_dict[word] += 1
else:
all_words_dict[word] = 1
all_words_tuple_list = sorted(all_words_dict.items(), key=lambda f: f[1], reverse=True) # 按值降序排序
all_words_list = list(list(zip(*all_words_tuple_list))[0]) # all_words_list[word_one, word_two, ...]
return all_words_list, train_data_list, test_data_list, train_class_list, test_class_list
# 选取特征词
def words_dict(all_words_list, deleteN, stopwords_set=set()):
feature_words = []
n = 1
for t in range(deleteN, len(all_words_list), 1):
if n > 1000: # 维度最大1000
break
# 非数字 非停用词 长度 1-4 之间
if not all_words_list[t].isdigit() and all_words_list[t] not in stopwords_set and 1 < len(all_words_list[t]) < 5:
feature_words.append(all_words_list[t])
n += 1
return feature_words
# 文本特征
def text_features(train_data_list, test_data_list, feature_words):
def text_feature_(text, feature_words):
text_words = set(text)
features = [1 if word in text_words else 0 for word in feature_words]
return features
train_feature_list = [text_feature_(text, feature_words) for text in train_data_list]
test_feature_list = [text_feature_(text, feature_words) for text in test_data_list]
return train_feature_list, test_feature_list
# 对停用词去重
def make_word_set(words_file):
words_set = set()
with open(words_file, 'r', encoding='UTF-8') as fp:
for line in fp.readlines():
word = line.strip()
if len(word)>0 and word not in words_set:
words_set.add(word)
return words_set
# 列表求均值
def average(accuracy_list):
sum = 0
for i in accuracy_list:
sum += i
return round(sum/len(accuracy_list),3)
# 分类 同时输出准确率等
def text_classifier(train_feature_list, test_feature_list, train_class_list, test_class_list, flag):
if flag == 'nb':
# 朴素贝叶斯分类器 拟合 默认拉普拉斯平滑 不指定先验概率先验概率
classifier = NB().fit(train_feature_list, train_class_list)
if flag == 'lg':
# 逻辑回归分类器 指定liblinear为求解最优化问题的算法 最大迭代数 多分类问题策略
classifier = LR(solver='liblinear',max_iter=5000, multi_class='auto').fit(train_feature_list, train_class_list)
if flag == 'rf':
# 随机森林分类器
classifier = RF(n_estimators=200).fit(train_feature_list, train_class_list)
test_accuracy = classifier.score(test_feature_list, test_class_list) # 测试准确率
return test_accuracy
def start(flag):
folder_path = 'D:/WorkSpace/THUCTC/THUCNews/' # 请修改成自己的路径
all_words_list, train_data_list, test_data_list, train_class_list, test_class_list = text_processor(folder_path, test_size=0.2)
stopwords_set = make_word_set('D:/WorkSpace/tmp/py/stop_words_cn.txt')
# 文本特征的提取和分类
deleteNs = range(0,1000,20)
test_accuracy_list = []
# 每循环一次，去除前 20 个最高词频，直到去除 980 个最高词频为止
for deleteN in deleteNs:
feature_words = words_dict(all_words_list, deleteN, stopwords_set)
train_feature_list, test_feature_list = text_features(train_data_list, test_data_list, feature_words)
if flag == 'nb':
test_accuracy = text_classifier(train_feature_list, test_feature_list, train_class_list, test_class_list, flag='nb')
if flag == 'lg':
test_accuracy = text_classifier(train_feature_list, test_feature_list, train_class_list, test_class_list, flag='lg')
if flag == 'rf':
test_accuracy = text_classifier(train_feature_list, test_feature_list, train_class_list, test_class_list, flag='rf')
test_accuracy_list.append(test_accuracy)
print(flag + '平均准确度：', average(test_accuracy_list))
print(flag + '最大准确度：', round(max(test_accuracy_list), 3))
return deleteNs, test_accuracy_list
if __name__ == "__main__":
plt.figure(figsize=(13, 11))
for i in range(5):
# 1
flag = 'nb'
nb_deleteNs, nb_accuracy_list = start(flag)
flag = 'lg'
lg_deleteNs, lg_accuracy_list = start(flag)
flag = 'rf'
rf_deleteNs, rf_accuracy_list = start(flag)
# 绘图
plt.title('Relationship of deleteNs and test_accuracy')
plt.xlabel('deleteNs')
plt.ylabel('test_accuracy')
plt.grid()
plt.plot(nb_deleteNs, nb_accuracy_list, 'b', label='nb')
plt.plot(lg_deleteNs, lg_accuracy_list, 'k', label='lg')
plt.plot(rf_deleteNs, rf_accuracy_list, 'r', label='rf')
plt.annotate('大', xy=((nb_accuracy_list.index(max(nb_accuracy_list))-1)*20, max(nb_accuracy_list)))
plt.annotate('大', xy=((lg_accuracy_list.index(max(lg_accuracy_list))-1)*20, max(lg_accuracy_list)))
plt.annotate('大', xy=((rf_accuracy_list.index(max(rf_accuracy_list))-1)*20, max(rf_accuracy_list)))
plt.legend()
plt.show()
运行结果：
其他参数请自行修改
