Detecting Insults in Social Commentary 数据分析报告（python）

Detecting Insults in Social Commentary 数据分析报告

报告摘要

• 目标：本分析旨在利用文本数据判断一个评论是否为侮辱性评论。
• 方法：对评论数据进行数据清洗、停止词处理、词干化基础上，构建词频矩阵，利用逻辑斯蒂回归和L1正则化的逻辑回归对评论是否为侮辱性评论进行判断。
• 结论：对测试集数据进行测试后，发现模型具有一定的判断效果。

目录

• 问题描述
• 数据加载
• 文本数据处理
  • 数据清洗
  • 停止词处理
  • 文本词干化处理
  • 计算词频矩阵
• 模型构建与评估
  • 划分训练集和测试集
  • 利用逻辑斯蒂模型建模
  • L1正则化建模

一、问题描述

本问题旨在判断一个评论是否为侮辱性评论。每个样本由一句/一段评论构成，判断每个评论是否为针对个人的侮辱性评论。

变量名	含义
Insult	评论是否为侮辱性评论
Date	评论时间
Comment	评论内容

二、数据加载

# 加载所需的python库
import statsmodels.api as sm
import statsmodels.formula.api as smf
import statsmodels.graphics.api as smg
import patsy
%matplotlib inline
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from pandas import Series,DataFrame
from scipy import stats
import seaborn as sns

#载入train数据集
traindata = pd.read_csv("D:/学习/数据挖掘与机器学习/homework week3/train.csv")
train = traindata

# 查看前五条数据
train.head()

	Insult	Date	Comment
0	1	20120618192155Z	"You fuck your dad."
1	0	20120528192215Z	"i really don't understand your point.\xa0 It ...
2	0	NaN	"A\\xc2\\xa0majority of Canadians can and has ...
3	0	NaN	"listen if you dont wanna get married to a man...
4	0	20120619094753Z	"C\xe1c b\u1ea1n xu\u1ed1ng \u0111\u01b0\u1edd...

# 训练集中共有3947条数据，其中Insult和Comment没有缺失值，Date有718条缺失。
train.info()

Int64Index: 3947 entries, 0 to 3946
Data columns (total 3 columns):
Insult     3947 non-null int64
Date       3229 non-null object
Comment    3947 non-null object
dtypes: int64(1), object(2)
memory usage: 123.3+ KB

三、文本数据处理

3.1 数据清洗

# 构建数据清洗函数、去掉标点等符号
import re
def preprocessor(text):
    text = re.sub('<[^>]*>', '', text)
    emoticons = re.findall('(?::|;|=)(?:-)?(?:\)|\(|D|P)', text)
    text = re.sub('[\W]+', ' ', text.lower()) + \
           ' '.join(emoticons).replace('-', '')
    return text

# 利用构建的函数进行数据清洗
train.Comment = train.Comment.apply(preprocessor)

train.Comment[1]

' i really don t understand your point xa0 it seems that you are mixing apples and oranges '

3.2 停止词处理

# 载入停止词库
import nltk
nltk.download('stopwords')

[nltk_data] Downloading package stopwords to
[nltk_data]     C:\Users\yunlai\AppData\Roaming\nltk_data...
[nltk_data]   Package stopwords is already up-to-date!





True

# 去掉停止词
from nltk.corpus import stopwords
stop = stopwords.words('english')
[w for w in train.Comment if w not in stop]
train.Comment.head()

0                                   you fuck your dad 
1     i really don t understand your point xa0 it s...
2     a xc2 xa0majority of canadians can and has be...
3     listen if you dont wanna get married to a man...
4     c xe1c b u1ea1n xu u1ed1ng u0111 u01b0 u1eddn...
Name: Comment, dtype: object

3.3 文本词干化处理

from nltk.stem.porter import PorterStemmer
porter = PorterStemmer()
def tokenizer_porter(text):
    return [porter.stem(word) for word in text.split()]

# 词干化
train.Comment = train.Comment.apply(tokenizer_porter)

train.Comment[0]

['you', 'fuck', 'your', 'dad']

# 编写函数、将词干化后的词连接
def join_data(text):
    text = ' '.join(text)
    return text

train.Comment = train.Comment.apply(join_data)

train.Comment[1]

'i realli don t understand your point xa0 it seem that you are mix appl and orang'

3.4 计算词频矩阵

from sklearn.feature_extraction.text import TfidfTransformer
tfidf = TfidfTransformer(use_idf=True, norm='l2', smooth_idf=True)
from sklearn.feature_extraction.text import CountVectorizer
count = CountVectorizer()

# 计算每个词的词频矩阵
comment = tfidf.fit_transform(count.fit_transform(train.Comment)).toarray()           

comment =DataFrame(comment)

# 将计算结果合并到数据集中
train = pd.merge(train,comment,left_index = True, right_index = True)

四、模型构建与评估

4.1 划分训练集和测试集数据

data = train

del data["Date"]
del data["Comment"]

data.head()

	Insult	0	1	2	3	4	5	6	7	8	...	12691	12692	12693	12694	12695	12696	12697	12698	12699	12700
0	1	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0
1	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0
2	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0
3	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0
4	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0

5 rows × 12702 columns

from sklearn.linear_model import LogisticRegression
from sklearn.metrics import confusion_matrix, roc_curve,roc_auc_score,classification_report 
from sklearn.cross_validation import cross_val_score
from sklearn.cross_validation import train_test_split

train_y = data.Insult
train_x = data
del train_x["Insult"]
train_x['intercept'] = 1.0

train_x.head()

	0	1	2	3	4	5	6	7	8	9	...	12692	12693	12694	12695	12696	12697	12698	12699	12700	intercept
0	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	1
1	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	1
2	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	1
3	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	1
4	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	1

5 rows × 12702 columns

# 划分训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(
         train_x, train_y, test_size=0.3, random_state=0)

4.2 利用逻辑斯蒂模型建模

# 考虑到样本数据量较少，构建随机森林等模型效果可能不好，故构建逻辑斯蒂模型
clf = LogisticRegression()
clf.fit(X_train,y_train)

LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True,
          intercept_scaling=1, max_iter=100, multi_class='ovr', n_jobs=1,
          penalty='l2', random_state=None, solver='liblinear', tol=0.0001,
          verbose=0, warm_start=False)

# 利用模型进行预测
clf.predict(X_test)

array([0, 0, 0, ..., 0, 0, 1], dtype=int64)

preds = clf.predict(X_test)

# 计算混淆矩阵
confusion_matrix(y_test,preds)

array([[840,  22],
       [197, 126]])

# 计算roc_auc得分
pre = clf.predict_proba(X_test)
roc_auc_score(y_test,pre[:,1])

0.89364858166981531

# 画出roc曲线
fpr,tpr,thresholds = roc_curve(y_test,pre[:,1])
fig,ax = plt.subplots(figsize=(8,5))
plt.plot(fpr,tpr)
ax.set_title("Roc of Logistic Regression")

4.3 利用L1正则化建模

# 参数调整，C=2
lrtrain = LogisticRegression(penalty='l1', C=2)
lrtrain.fit(X_train,y_train)

LogisticRegression(C=2, class_weight=None, dual=False, fit_intercept=True,
          intercept_scaling=1, max_iter=100, multi_class='ovr', n_jobs=1,
          penalty='l1', random_state=None, solver='liblinear', tol=0.0001,
          verbose=0, warm_start=False)

# 利用模型预测，构建混淆矩阵
preds2 = lrtrain.predict(X_test)
confusion_matrix(y_test,preds2)

array([[809,  53],
       [146, 177]])

# 计算roc_auc得分
pre2 = lrtrain.predict_proba(X_test)
roc_auc_score(y_test,pre2[:,1])

0.8920287616817395

# 画出roc曲线
fpr,tpr,thresholds = roc_curve(y_test,pre2[:,1])
fig,ax = plt.subplots(figsize=(8,5))
plt.plot(fpr,tpr)
ax.set_title("Roc of Logistic Regression L1")

逻辑斯蒂模型和L1正则化的逻辑斯蒂模型roc_auc 得分分别为0.87和0.89，说明模型具有一定效果。