import os
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from matplotlib.pylab import style #自定义图表风格
style.use('ggplot')
from IPython.core.interactiveshell import InteractiveShell
InteractiveShell.ast_node_interactivity = "all"
plt.rcParams['font.sans-serif'] = ['Simhei'] # 解决中文乱码问题
import re
import jieba.posseg as psg
import itertools
#conda install -c anaconda gensim
from gensim import corpora,models #主题挖掘，提取关键信息
# pip install wordcloud
from wordcloud import WordCloud,ImageColorGenerator
from collections import Counter
from sklearn import tree
from sklearn.model_selection import train_test_split
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.metrics import classification_report
from sklearn.metrics import accuracy_score
import graphviz
raw_data=pd.read_csv('reviews.csv')
#raw_data.info()
#raw_data.columns
#取值分布
for cate in ['creationTime', 'nickname', 'referenceName', 'content_type']:
raw_data[cate].value_counts()
reviews=raw_data.copy()
reviews=reviews[['content', 'content_type']]
#print('去重之前：',reviews.shape[0])
reviews=reviews.drop_duplicates()
#print('去重之后：',reviews.shape[0])
# 清洗之前
content=reviews['content']
#print('清洗之前''-----------')
#for i in range(5,10):
# print(content[i])
# print('-----------')
#清洗之后，将数字、字母、京东美的电热水器字样都删除
info=re.compile('[0-9a-zA-Z]|京东|美的|电热水器|热水器|')
content=content.apply(lambda x: info.sub('',x)) #替换所有匹配项
#print('清洗之后''-----------')
#for i in range(5,10):
#print(content[i])
#print('-----------')
#分词，由元组组成的list
seg_content=content.apply( lambda s: [(x.word,x.flag) for x in psg.cut(s)] )
seg_content.shape
len(seg_content)
#print(seg_content[5])
#统计评论词数
n_word=seg_content.apply(lambda s: len(s))
len(n_word)
n_word.head(6)
#得到各分词在第几条评论
n_content=[ [x+1]*y for x,y in zip(list(seg_content.index),list(n_word))] #[x+1]*y,表示复制y份，由list组成的list
index_content_long=sum(n_content,[]) #表示去掉[]，拉平,返回list
len(index_content_long)
sum([[2,2],[3,3,3]],[])
#分词及词性，去掉[]，拉平
seg_content.head()
seg_content_long=sum(seg_content,[])
seg_content_long
type(seg_content_long)
len(seg_content_long)
seg_content_long[0]
#得到加长版的分词、词性
word_long=[x[0] for x in seg_content_long]
nature_long=[x[1] for x in seg_content_long]
len(word_long)
len(nature_long)
#content_type拉长
n_content_type=[ [x]*y for x,y in zip(list(reviews['content_type']),list(n_word))] #[x+1]*y,表示复制y份
content_type_long=sum(n_content_type,[]) #表示去掉[]，拉平
len(content_type_long)
review_long=pd.DataFrame({'index_content':index_content_long,
'word':word_long,
'nature':nature_long,
'content_type':content_type_long})
review_long.shape
review_long.head()
review_long['nature'].unique()
#print(review_long)
#去除标点符号
review_long_clean=review_long[review_long['nature']!='x'] #x表示标点符合
review_long_clean.shape
#导入停用词
stop_path=open('stoplist.txt','r',encoding='UTF-8')
stop_words=stop_path.readlines()
len(stop_words)
stop_words[0:5]
#停用词，预处理
stop_words=[word.strip('n') for word in stop_words]
stop_words[0:5]
#得到不含停用词的分词表
word_long_clean=list(set(word_long)-set(stop_words))
#print(len(word_long_clean))
review_long_clean=review_long_clean[review_long_clean['word'].isin(word_long_clean)]
#print(review_long_clean.shape)
#再次统计每条评论的分词数量
n_word=review_long_clean.groupby('index_content').count()['word']
n_word
index_word=[ list(np.arange(1,x+1)) for x in list(n_word)]
index_word_long=sum(index_word,[]) #表示去掉[]，拉平
len(index_word_long)
review_long_clean['index_word']=index_word_long
review_long_clean.head()
review_long_clean.to_csv('1_review_long_clean.csv')
n_review_long_clean=review_long_clean[[ 'n' in nat for nat in review_long_clean.nature]]
n_review_long_clean.shape
#print(n_review_long_clean.head())
#print(n_review_long_clean.nature.value_counts())
n_review_long_clean.to_csv('1_n_review_long_clean.csv')
#font=r"C:WindowsFontsmsyh.ttc"
font=r"C:WindowsFontsmsyh.ttc"
background_image=plt.imread('1.png')
wordcloud = WordCloud(font_path=font, max_words = 100, mode='RGBA' ,background_color='white',mask=background_image)
#wordcloud = WordCloud(font_path=font, max_words = 100, background_color='white',mask=background_image) #width=1600,height=1200, mode='RGBA'
#wordcloud = WordCloud(font_path=None,max_words = 100, background_color='white', mode='RGBA')
#print("okok+"+review_long_clean.word.values)
wordcloud.generate_from_frequencies(Counter(review_long_clean.word.values))
wordcloud.to_file('1_分词后的词云图.png')
plt.figure(figsize=(20,10))
plt.imshow(wordcloud)
plt.axis('off')
#plt.show()
font=r"C:WindowsFontsmsyh.ttc"
background_image=plt.imread('1.png')
wordcloud = WordCloud(font_path=font, max_words = 100, mode='RGBA' ,background_color='white',mask=background_image) #width=1600,height=1200
wordcloud.generate_from_frequencies(Counter(n_review_long_clean.word.values))
wordcloud.to_file('1_分词后的词云图(名词）.png')
plt.figure(figsize=(20,10))
plt.imshow(wordcloud)
plt.axis('off')
#plt.show()
#第一步：构造特征空间和标签
Y=[]
for ind in review_long_clean.index_content.unique():
y=[ word for word in review_long_clean.content_type[review_long_clean.index_content==ind].unique() ]
Y.append(y)
#print('len(Y)=',len(Y))
X=[]
for ind in review_long_clean.index_content.unique():
term=[ word for word in review_long_clean.word[review_long_clean.index_content==ind].values ]
X.append(' '.join(term))
#print('len(X)=',len(X))
#print(X)
#print(Y)
#第二步：训练集、测试集划分
x_train,x_test,y_train,y_test=train_test_split(X,Y,test_size=0.2,random_state=7)
#第三步：词转向量，01矩阵
count_vec=CountVectorizer(binary=True)
x_train=count_vec.fit_transform(x_train)
x_test=count_vec.transform(x_test)
#第四步：构建决策树
dtc=tree.DecisionTreeClassifier(max_depth=5)
dtc.fit(x_train,y_train)
#print('在训练集上的准确率：%.2f'% accuracy_score(y_train,dtc.predict(x_train)))
y_true=y_test
y_pred=dtc.predict(x_test)
#print(classification_report(y_true,y_pred))
#print('在测试集上的准确率：%.2f'% accuracy_score(y_true,y_pred))
#第五步：画决策树
#这个决策数图，需要在网上下载graphviz 包，并且配置环境变量
from graphviz import Digraph
import os
os.environ["PATH"] += os.pathsep + 'C:\Program Files\Graphviz\bin'
dot_data=tree.export_graphviz(dtc ,feature_names=count_vec.get_feature_names())
graph=graphviz.Source(dot_data)
#print(graph)
#graph.render("test", view=True)
#导入评价情感词
pos_comment=pd.read_csv('正面评价词语（中文）.txt',header=None,sep='/n',engine='python')
neg_comment=pd.read_csv('负面评价词语（中文）.txt',header=None,sep='/n',engine='python')
pos_emotion=pd.read_csv('正面情感词语（中文）.txt',header=None,sep='/n',engine='python')
neg_emotion=pd.read_csv('负面情感词语（中文）.txt',header=None,sep='/n',engine='python')
#print(pos_comment.shape)
#print(neg_comment.shape)
#print(pos_emotion.shape)
#print(neg_emotion.shape)
pos=pd.concat([pos_comment,pos_emotion],axis=0)
pos.shape
neg=pd.concat([neg_comment,neg_emotion],axis=0)
neg.shape
#增加新词
c='点赞'
c in pos.values
d='歇菜'
d in neg.values
new_pos=pd.Series(['点赞'])
new_neg=pd.Series(['歇菜'])
positive=pd.concat([pos,new_pos],axis=0)
positive.shape
negative=pd.concat([neg,new_neg],axis=0)
negative.shape
positive.columns=['review']
positive['weight']=pd.Series([1]*len(positive))
#print(positive)
negative.columns=['review']
negative['weight']=pd.Series([-1]*len(negative))
#print(negative)
pos_neg=pd.concat([positive,negative],axis=0)
pos_neg.shape
#合并到review_long_clean中
#表联接
data=review_long_clean.copy()
review_mltype=pd.merge(data,pos_neg,how='left',left_on='word',right_on='review')
review_mltype.shape
review_mltype=review_mltype.drop(['review'],axis=1)
review_mltype=review_mltype.replace(np.nan,0)
#print(review_mltype.head())
#修正情感倾向
notdict=pd.read_csv('not.csv')
#print(notdict.shape)
notdict['freq']=[1]*len(notdict)
#print(notdict)
#准备一
review_mltype['amend_weight']=review_mltype['weight']
review_mltype['id']=np.arange(0,review_mltype.shape[0])
#print(review_mltype)
# 准备二，只保留有情感值的行
only_review_mltype=review_mltype[review_mltype['weight']!=0]
only_review_mltype.index=np.arange(0,only_review_mltype.shape[0]) #索引重置
only_review_mltype.shape
#print(only_review_mltype)
i=4
review_i=review_mltype[review_mltype['index_content']==only_review_mltype['index_content'][i]]
#print(review_i)#第i个情感词的评论
# 看该情感词前2个词，来判罚否定的语气。如果在句首，则没有否词，如果在句子的第二次词，则看前1个词，来判断否定的语气。
index = only_review_mltype['id']
for i in range(0, only_review_mltype.shape[0]):
review_i = review_mltype[review_mltype['index_content'] == only_review_mltype['index_content'][i]] # 第i个情感词的评论
review_i.index = np.arange(0, review_i.shape[0]) # 重置索引后，索引值等价于index_word
word_ind = only_review_mltype['index_word'][i] # 第i个情感值在该条评论的位置
# 第一种，在句首。则不用判断
# 第二种，在评论的第2个为位置
if word_ind == 2:
ne = sum([review_i['word'][word_ind - 1] in notdict['term']])
if ne == 1:
review_mltype['amend_weight'][index[i]] = -(review_mltype['weight'][index[i]])
# 第三种，在评论的第2个位置以后
elif word_ind > 2:
ne = sum([word in notdict['term'] for word in
review_i['word'][[word_ind - 1, word_ind - 2]]]) # 注意用中括号[word_ind-1,word_ind-2]
if ne == 1:
review_mltype['amend_weight'][index[i]] = - (review_mltype['weight'][index[i]])
# print(review_mltype.shape)
review_mltype[(review_mltype['weight'] - review_mltype['amend_weight']) != 0] # 说明两列值一样
#计算每条评论的情感值
#print(review_mltype.tail())
emotion_value = review_mltype.groupby('index_content', as_index=False)['amend_weight'].sum()
#print(emotion_value)
emotion_value.to_csv('./1_emotion_value', index=True, header=True)
#查看情感分析效果
# 每条评论的amend_weight总和不等于零
content_emotion_value = emotion_value.copy()
content_emotion_value.shape
content_emotion_value = content_emotion_value[content_emotion_value['amend_weight'] != 0]
content_emotion_value['ml_type'] = ''
content_emotion_value['ml_type'][content_emotion_value['amend_weight'] > 0] = 'pos'
content_emotion_value['ml_type'][content_emotion_value['amend_weight'] < 0] = 'neg'
content_emotion_value.shape
print(content_emotion_value)
# 每条评论的amend_weight总和等于零
# 这个方法其实不好用，有一半以上的评论区分不出正、负情感。
content_emotion_value0 = emotion_value.copy()
content_emotion_value0 = content_emotion_value0[content_emotion_value0['amend_weight'] == 0]
content_emotion_value0.head()
print(raw_data.content[6])
print(raw_data.content[7])
print(raw_data.content[8])
# 合并到大表中
content_emotion_value = content_emotion_value.drop(['amend_weight'], axis=1)
review_mltype.shape
review_mltype = pd.merge(review_mltype, content_emotion_value, how='left', left_on='index_content',
right_on='index_content')
review_mltype = review_mltype.drop(['id'], axis=1)
review_mltype.shape
print(review_mltype)
review_mltype.to_csv('./1_review_mltype', index=True, header=True)
cate = ['index_content', 'content_type', 'ml_type']
data_type = review_mltype[cate].drop_duplicates()
confusion_matrix = pd.crosstab(data_type['content_type'], data_type['ml_type'], margins=True)
print(confusion_matrix)
data = data_type[['content_type', 'ml_type']]
data = data.dropna(axis=0)
print(classification_report(data['content_type'], data['ml_type']))
data = review_mltype.copy()
data = data[data['amend_weight'] != 0]
word_data_pos = data[data['ml_type'] == 'pos']
word_data_neg = data[data['ml_type'] == 'neg']
# 按照以上修改，显示信息
font = r"C:WindowsFontsmsyh.ttc"
background_image = plt.imread('1.png')
wordcloud = WordCloud(font_path=font, max_words=100, mode='RGBA', background_color='white',
mask=background_image) # width=1600,height=1200
# wordcloud = WordCloud(max_words = 100, mode='RGBA' ,background_color='white') #width=1600,height=1200
wordcloud.generate_from_frequencies(Counter(word_data_pos.word.values))
plt.figure(figsize=(15, 7))
plt.imshow(wordcloud)
plt.axis('off')
#plt.show()
# font=r"C:WindowsFontsmsyh.ttc"
font = r"C:WindowsFontsmsyh.ttc"
background_image = plt.imread('1.png')
# background_image=plt.imread('./p6sad.jpg')
wordcloud = WordCloud(font_path=font, max_words=100, mode='RGBA', background_color='white',
mask=background_image) # width=1600,height=1200
# wordcloud = WordCloud(max_words = 100, mode='RGBA' ,background_color='white') #width=1600,height=1200
wordcloud.generate_from_frequencies(Counter(word_data_neg.word.values))
plt.figure(figsize=(15, 7))
plt.imshow(wordcloud)
plt.axis('off')
#plt.show()
data = review_mltype.copy()
word_data_pos = data[data['ml_type'] == 'pos']
word_data_neg = data[data['ml_type'] == 'neg']
font = r"C:WindowsFontsmsyh.ttc"
background_image = plt.imread('1.png')
wordcloud = WordCloud(font_path=font, max_words=100, mode='RGBA', background_color='white',
mask=background_image) # width=1600,height=1200
wordcloud.generate_from_frequencies(Counter(word_data_pos.word.values))
plt.figure(figsize=(15, 7))
plt.imshow(wordcloud)
plt.axis('off')
#plt.show()
background_image = plt.imread('1.png')
wordcloud = WordCloud(font_path=font, max_words=100, mode='RGBA', background_color='white',
mask=background_image) # width=1600,height=1200
wordcloud.generate_from_frequencies(Counter(word_data_neg.word.values))
plt.figure(figsize=(15, 7))
plt.imshow(wordcloud)
plt.axis('off')
#plt.show()
data = review_mltype.copy()
word_data_pos = data[data['ml_type'] == 'pos']
word_data_neg = data[data['ml_type'] == 'neg']
# 建立词典，去重
pos_dict = corpora.Dictionary([[i] for i in word_data_pos.word]) # shape=(n,1)
neg_dict = corpora.Dictionary([[i] for i in word_data_neg.word])
#print(pos_dict)
# 建立语料库
pos_corpus = [pos_dict.doc2bow(j) for j in [[i] for i in word_data_pos.word]] # shape=(n,(2,1))
neg_corpus = [neg_dict.doc2bow(j) for j in [[i] for i in word_data_neg.word]]
len(word_data_pos.word)
len(pos_dict)
len(pos_corpus)
#print(pos_corpus) # 元素是元组，元组（x,y),x是在词典中的位置，y是1表示存在。
# 构造主题数寻优函数
def cos(vector1, vector2):
'''
函数功能：余玄相似度函数
'''
dot_product = 0.0
normA = 0.0
normB = 0.0
for a, b in zip(vector1, vector2):
dot_product += a * b
normA += a ** 2
normB += b ** 2
if normA == 0.0 or normB == 0.0:
return None
else:
return (dot_product / ((normA * normB) ** 0.5))
# 主题数寻优
# 这个函数可以重复调用，解决其他项目的问题
def LDA_k(x_corpus, x_dict):
'''
函数功能：
'''
# 初始化平均余玄相似度
mean_similarity = []
mean_similarity.append(1)
# 循环生成主题并计算主题间相似度
for i in np.arange(2, 11):
lda = models.LdaModel(x_corpus, num_topics=i, id2word=x_dict) # LDA模型训练
for j in np.arange(i):
term = lda.show_topics(num_words=50)
# 提取各主题词
top_word = [] # shape=(i,50)
for k in np.arange(i):
top_word.append([''.join(re.findall('"(.*)"', i)) for i in term[k][1].split('+')]) # 列出所有词
# 构造词频向量
word = sum(top_word, []) # 列车所有词
unique_word = set(word) # 去重
# 构造主题词列表，行表示主题号，列表示各主题词
mat = [] # shape=(i,len(unique_word))
for j in np.arange(i):
top_w = top_word[j]
mat.append(tuple([top_w.count(k) for k in unique_word])) # 统计list中元素的频次，返回元组
# 两两组合。方法一
p = list(itertools.permutations(list(np.arange(i)), 2)) # 返回可迭代对象的所有数学全排列方式。
y = len(p) # y=i*(i-1)
top_similarity = [0]
for w in np.arange(y):
vector1 = mat[p[w][0]]
vector2 = mat[p[w][1]]
top_similarity.append(cos(vector1, vector2))
# #两两组合，方法二
# for x in range(i-1):
# for y in range(x,i):
# 计算平均余玄相似度
mean_similarity.append(sum(top_similarity) / y)
return mean_similarity
# 计算主题平均余玄相似度
pos_k = LDA_k(pos_corpus, pos_dict)
neg_k = LDA_k(neg_corpus, neg_dict)
pos_k
neg_k
pd.Series(pos_k, index=range(1, 11)).plot()
plt.title('正面评论LDA主题数寻优')
plt.show()
pd.Series(neg_k, index=range(1, 11)).plot()
plt.title('负面评论LDA主题数寻优')
plt.show()
pos_lda = models.LdaModel(pos_corpus, num_topics=2, id2word=pos_dict)
neg_lda = models.LdaModel(neg_corpus, num_topics=2, id2word=neg_dict)
pos_lda.print_topics(num_topics=10)
neg_lda.print_topics(num_topics=10)
