在 TF-IDF 特征提取的基础上对模型建立与评估

Question

在 TF-IDF 特征提取的基础上， 分别采用逻辑回归、 朴素贝叶斯两种模型对文本数据的分类效果进行比较（准确率、召回率、F1）；逻辑回归对分类标签种类比较（准确率、召回率、F1）
数据来源：https://github.com/aceimnorstuvwxz/toutiao-text-classfication-dataset

import numpy as np
import pandas as pd
import time
import jieba
import re
import string
import pickle
import matplotlib.pyplot as plt


from sklearn.metrics import accuracy_score, precision_score, recall_score
from sklearn.feature_extraction.text import TfidfTransformer, CountVectorizer
from sklearn.model_selection import StratifiedKFold



# # Pandas设置
# pd.set_option("display.max_columns", None)  # 设置显示完整的列
# pd.set_option("display.max_rows", None)  # 设置显示完整的行
# pd.set_option("display.expand_frame_repr", False)  # 设置不折叠数据
# pd.set_option("display.max_colwidth", 100)  # 设置列的最大宽度


# 加载数据集
def data():
    a=[]
    with open(r'D:\桌面\文本分类\data.txt',encoding='utf-8') as f:
        a.append(f.readlines())
    df = pd.DataFrame(a[0])
    df['label']=df[0].apply(lambda x:x.split('_!_')[1])      #对第二个！前的数据名称设置为label
    df['label_desc']=df[0].apply(lambda x:x.split('_!_')[2]) #对第三个！前的数据名称设置为label_desc
    df['sentence']=df[0].apply(lambda x:x.split('_!_')[3])   #对第四个！前的数据名称设置为sentence
    return df.iloc[0:,1:] 
df=data()


#去重仅保留第一次出现的评论
df=df.drop_duplicates(keep='first',inplace=False)
#去除空白行
df= df.dropna(axis=0)



STOPWORDS = r'D:\桌面\文本分类\停用词\stopwords-master\哈工大停用词表.txt'
stoplist = []
content_list = []
for word in open(STOPWORDS,encoding='utf-8'):
    stoplist.append(word.strip())   #添加停用词表


#去除无效符号
def qu_dian(text):
    punc = r'[~`!#$%^&*()_+-=|\';":/.,?><~·！@#￥%……&*（）——+-=“：’；、。，？》《{}]'
    ff=re.sub(punc, "",text)
    f1=jieba.lcut(ff,cut_all=False)     #中文分词
    def rn(x):
        return x.replace(' ','')
    ben=list(filter(rn,f1))
    return ben

df['quchu']=df['sentence'].apply(qu_dian)

#去除停用词
def qu_word(x):
    real=[]
    for i in x:
        if i not in stoplist:
            real.append(i)
    return real

df['quchu']=df['quchu'].apply(qu_word)



# 实例化CountVectorizer
vectorizer = CountVectorizer()
# 将文本数据转化为特征矩阵
X = vectorizer.fit_transform(df['sentence'].tolist())
# 实例化TfidfTransformer
transformer = TfidfTransformer()
# 对特征矩阵进行tf-idf转换
X = transformer.fit_transform(X)
# 实例化StratifiedKFold
skf = StratifiedKFold(n_splits=5)

# 对数据进行划分
for train_index, test_index in skf.split(X, df['label']):
    X_train, X_test = X[train_index], X[test_index]
    y_train, y_test = df['label'][train_index], df['label'][test_index]

Answer 1

import numpy as np
import pandas as pd
import time
import jieba
import re
import string
import pickle
import matplotlib.pyplot as plt


from sklearn.metrics import accuracy_score, precision_score, recall_score
from sklearn.feature_extraction.text import TfidfTransformer, CountVectorizer
from sklearn.model_selection import StratifiedKFold
from sklearn.metrics import classification_report
from sklearn.linear_model import LogisticRegression
from sklearn.naive_bayes import MultinomialNB


# # Pandas设置
# pd.set_option("display.max_columns", None)  # 设置显示完整的列
# pd.set_option("display.max_rows", None)  # 设置显示完整的行
# pd.set_option("display.expand_frame_repr", False)  # 设置不折叠数据
# pd.set_option("display.max_colwidth", 100)  # 设置列的最大宽度


# 加载数据集
def data():
    a=[]
    with open(r'F:\项目\jupternotebook\吴恩达机器学习2022\文本数据分析\datas.txt',encoding='utf-8') as f:
        a.append(f.readlines())
    df = pd.DataFrame(a[0])
    df['label']=df[0].apply(lambda x:x.split('_!_')[1])      #对第二个！前的数据名称设置为label
    df['label_desc']=df[0].apply(lambda x:x.split('_!_')[2]) #对第三个！前的数据名称设置为label_desc
    df['sentence']=df[0].apply(lambda x:x.split('_!_')[3])   #对第四个！前的数据名称设置为sentence
    return df.iloc[0:,1:] 
df=data()


#去重仅保留第一次出现的评论
df=df.drop_duplicates(keep='first',inplace=False)
#去除空白行
df= df.dropna(axis=0)



STOPWORDS = r'F:\项目\jupternotebook\吴恩达机器学习2022\文本数据分析\哈工大停用词表.txt'
stoplist = []
content_list = []
for word in open(STOPWORDS,encoding='utf-8'):
    stoplist.append(word.strip())   #添加停用词表


#去除无效符号
def qu_dian(text):
    punc = r'[~`!#$%^&*()_+-=|\';":/.,?><~·！@#￥%……&*（）——+-=“：’；、。，？》《{}]'
    ff=re.sub(punc, "",text)
    f1=jieba.lcut(ff,cut_all=False)     #中文分词
    def rn(x):
        return x.replace(' ','')
    ben=list(filter(rn,f1))
    return ben

df['quchu']=df['sentence'].apply(qu_dian)

#去除停用词
def qu_word(x):
    real=[]
    for i in x:
        if i not in stoplist:
            real.append(i)
    return real

df['quchu']=df['quchu'].apply(qu_word)



# 实例化CountVectorizer
vectorizer = CountVectorizer()
# 将文本数据转化为特征矩阵
X = vectorizer.fit_transform(df['sentence'].tolist())
# 实例化TfidfTransformer
transformer = TfidfTransformer()
# 对特征矩阵进行tf-idf转换
X = transformer.fit_transform(X)
# 实例化StratifiedKFold
skf = StratifiedKFold(n_splits=5)

    
# 对数据进行划分
y = df['label']
for train_index, test_index in skf.split(X,y):
    X_train, X_test = X[train_index], X[test_index]
    y_train, y_test = y.iloc[train_index], y.iloc[test_index]
    
#逻辑回归

clf = LogisticRegression(random_state=0, solver='lbfgs', multi_class='multinomial')
clf.fit(X_train, y_train)
y_pred_log = clf.predict(X_test)

#计算准确率、召回率、F1
accuracy = accuracy_score(y_test, y_pred_log)
precision = precision_score(y_test, y_pred_log, average='macro')
recall = recall_score(y_test, y_pred_log, average='macro')
f1 = 2 * precision * recall / (precision + recall)
print("逻辑回归分类结果：\n准确率：", accuracy, "\n召回率：", recall, "\nF1：", f1)


#朴素贝叶斯

clf = MultinomialNB()
clf.fit(X_train, y_train)
y_pred_clf = clf.predict(X_test)

#计算准确率、召回率、F1
accuracy = accuracy_score(y_test, y_pred_clf)
precision = precision_score(y_test, y_pred_clf, average='macro')
recall = recall_score(y_test, y_pred_clf, average='macro')
f1 = 2 * precision * recall / (precision + recall)
print("朴素贝叶斯分类结果：\n准确率：", accuracy, "\n召回率：", recall, "\nF1：", f1)

# 评估分类效果
print(classification_report(y_test, y_pred_log))