第九章 分析文本数据和社交媒体
1 安装nltk 略
2 滤除停用字 姓名和数字
示例代码如下:
import nltk
# 加载英语停用字语料
sw = set(nltk.corpus.stopwords.words('english'))
print('Stop words', list(sw)[:7])
# 取得gutenberg语料库中的部分文件
gb = nltk.corpus.gutenberg
print('Gutenberg files', gb.fileids()[-5:])
# 取milton-paradise.txt文件中的前两句,作为下面所用的过滤语句
text_sent = gb.sents("milton-paradise.txt")[:2]
print('Unfiltered', text_sent)
# 过滤停用字
for sent in text_sent:
filtered = [w for w in sent if w.lower() not in sw]
print('Filtered', filtered)
# 取得文本内所含的标签
tagged = nltk.pos_tag(filtered)
print("Tagged", tagged)
words = []
for word in tagged:
if word[1] != 'NNP' and word[1] != 'CD':
words.append(word[0])
print(words)
# 词性标注集
# print(nltk.tag.tagset_mapping('ru-rnc', 'universal'))
运行结果如下:
Connected to pydev debugger (build162.1967.10)
Stop words ['his', 'only', 'because','with', 'each', 'myself', 'both']
Gutenberg files ['milton-paradise.txt','shakespeare-caesar.txt', 'shakespeare-hamlet.txt', 'shakespeare-macbeth.txt','whitman-leaves.txt']
Unfiltered [['[', 'Paradise', 'Lost', 'by','John', 'Milton', '1667', ']'], ['Book', 'I']]
Filtered ['[', 'Paradise', 'Lost', 'John','Milton', '1667', ']']
Tagged [('[', 'JJ'), ('Paradise', 'NNP'),('Lost', 'NNP'), ('John', 'NNP'), ('Milton', 'NNP'), ('1667', 'CD'), (']','NN')]
['[', ']']
Filtered ['Book']
Tagged [('Book', 'NN')]
['Book']
本例用到的标记集:
{'PRP$',
'PDT',
'CD',
'EX',
'.',
'NNS',
'MD',
'PRP',
'RP',
'(',
'VBD',
'``',
"''",
'NN', 名词
'LS',
'VBN',
'WRB',
'IN', 介词
'FW',
'POS',
'CC', 并连词
':',
'DT',
'VBZ',
'RBS',
'RBR',
'WP$',
'RB',
'SYM',
'JJS',
'JJR',
'UH',
'WDT',
'#',
',',
')',
'VB',
'NNPS',
'VBP', 动词
'NNP',
'JJ', 形容词
'WP',
'VBG',
'$',
'TO'} 单词to
粗略的分为以下12种类型
'VERB',
'NOUN',
'PRON',
'ADJ',
'ADV',
'ADP',
'CONJ',
'DET',
'NUM',
'PRT',
'X',
'.'
3 词袋模型
安装scikit-learn略
示例代码如下:
import nltk
from sklearn.feature_extraction.text import CountVectorizer
# 从gutenberg语料库中加载以下两个文件
gb = nltk.corpus.gutenberg
hamlet = gb.raw('shakespeare-hamlet.txt')
macbeth = gb.raw("shakespeare-macbeth.txt")
# 去掉英语停用词
cv = CountVectorizer(stop_words='english')
# 输出部分特征值
print("Feature vector", cv.fit_transform([hamlet, macbeth]).toarray())
# 特征值是按字母顺序排序
print('Features', cv.get_feature_names()[:5])
运行结果如下:
Feature vector [[ 1 0 1..., 14 0 1]
[0 1 0 ..., 1 1 0]]
Features ['1599', '1603', 'abhominably','abhorred', 'abide']
4 词频分析
示例代码如下:
def printLine(values, num, keyOrValue, tag):
"""
打印指定列表的num个元素的key或是value,输出标签为tag
:param values:列表
:param num: 输出元素个数
:param keyOrValue: 输出元素的键还是值 0表示键,1表示值
:param tag: 输出标签
:return:
"""
tmpValue = []
for key in sorted(values.items(), key=lambda d: d[1], reverse=True)[:num]:
tmpValue.append(key[keyOrValue])
print(tag, ":", tmpValue)
# 加载文档
gb = nltk.corpus.gutenberg
words = gb.words("shakespeare-caesar.txt")
# 支除停用词和标点符号
sw = set(nltk.corpus.stopwords.words('english'))
punctuation = set(string.punctuation)
filtered = [w.lower() for w in words if w.lower() not in sw and w.lower() not in punctuation]
# 创建freqDist对象,输出频率最高的键和值
fd = nltk.FreqDist(filtered)
printLine(fd, 5, 0, "Wrods")
printLine(fd, 5, 1, "Counts")
# 最常出现的单词和次数
print('Max', fd.max())
print('Count', fd['caesar'])
# 最常出现的双字词和词数
fd = nltk.FreqDist(nltk.bigrams(filtered))
printLine(fd, 5, 0, "Bigrams")
printLine(fd, 5, 1, "Counts")
print('Bigram Max', fd.max())
print('Bigram count', fd[('let', 'vs')])
# 最常出现的三字词和词数
fd = nltk.FreqDist(nltk.trigrams(filtered))
printLine(fd, 5, 0, "Trigrams")
printLine(fd, 5, 1, "Counts")
print('Bigram Max', fd.max())
print('Bigram count', fd[('enter', 'lucius', 'luc')])
运行结果如下:
Wrods : ['caesar', 'brutus', 'bru', 'haue','shall']
Counts : [190, 161, 153, 148, 125]
Max caesar
Count 190
Bigrams : [('let', 'vs'), ('wee', 'l'),('mark', 'antony'), ('marke', 'antony'), ('st', 'thou')]
Counts : [16, 15, 13, 12, 12]
Bigram Max ('let', 'vs')
Bigram count 16
Trigrams : [('enter', 'lucius', 'luc'),('wee', 'l', 'heare'), ('thee', 'thou', 'st'), ('beware', 'ides', 'march'),('let', 'vs', 'heare')]
Counts : [4, 4, 3, 3, 3]
Bigram Max ('enter', 'lucius', 'luc')
Bigram count 4
5 朴素贝叶斯分类
是一个概率算法,基于概率和数理统计中的贝叶斯定理
示例代码如下:
import nltk
import string
import random
# 停用词和标点符号集合
sw = set(nltk.corpus.stopwords.words('english'))
punctuation = set(string.punctuation)
# 将字长作为一个特征
def word_features(word):
return {'len': len(word)}
# 是否为停用词或是标点符号
def isStopword(word):
return word in sw or word in punctuation
# 加载文件
gb = nltk.corpus.gutenberg
words = gb.words("shakespeare-caesar.txt")
# 对单词进行标注,区分是否为停用词
labeled_words = ([(word.lower(), isStopword(word.lower())) for word in words])
random.seed(42)
random.shuffle(labeled_words)
print(labeled_words[:5])
# 求出每个单词的长度,作为特征值
featuresets = [(word_features(n), word) for (n, word) in labeled_words]
# 训练一个朴素贝叶斯分类器
cutoff = int(.9 * len(featuresets))
# 创建训练数据集和测试数据集
train_set, test_set = featuresets[:cutoff], featuresets[cutoff:]
# 检查分类器效果
classifier = nltk.NaiveBayesClassifier.train(train_set)
print("'behold' class", classifier.classify(word_features('behold')))
print("'the' class", classifier.classify(word_features('the')))
# 根据测试数据集来计算分类器的准确性
print("Accuracy", nltk.classify.accuracy(classifier, test_set))
# 贡献度最大的特征
print(classifier.show_most_informative_features(5))
运行结果如下:
[('i', True), ('is', True), ('in', True),('he', True), ('ambitious', False)]
'behold' class False
'the' class True
Accuracy 0.8521671826625387
Most Informative Features
len = 7 False : True = 77.8 : 1.0
len = 6 False : True = 52.2 : 1.0
len = 1 True : False = 51.8 : 1.0
len = 2 True : False = 10.9 : 1.0
len = 5 False : True = 10.9 : 1.0
None
6 情感分析
示例代码如下:
import random
from nltk.corpus import movie_reviews
from nltk.corpus import stopwords
from nltk import FreqDist
from nltk import NaiveBayesClassifier
from nltk.classify import accuracy
import string
def getElementsByNum(values, num, keyOrValue):
"""
取得指定列表的num个元素的key或是value,
:param values:列表
:param num: 元素个数
:param keyOrValue: 元素的键还是值 0表示键,1表示值
:return:
"""
tmpValue = []
for key in sorted(values.items(), key=lambda d: d[1], reverse=True)[:num]:
tmpValue.append(key[keyOrValue])
return tmpValue
# 加载数据
labeled_docs = [(list(movie_reviews.words(fid)), cat)
for cat in movie_reviews.categories()
for fid in movie_reviews.fileids(cat)]
random.seed(42)
random.shuffle(labeled_docs)
review_words = movie_reviews.words()
print("#Review Words", len(review_words))
# 设置停用词和标点符号
sw = set(stopwords.words('english'))
punctuation = set(string.punctuation)
# 检查是否为停用词
def isStopWord(word):
return word in sw or word in punctuation
# 过滤停用词和标点符号
filtered = [w.lower() for w in review_words if not isStopWord(w.lower())]
# print("# After filter", len(filtered))
# 选用词频最高的前5%作为特征
words = FreqDist(filtered)
N = int(.05 * len(words.keys()))
# word_features = words.keys()[:N]
word_features = getElementsByNum(words, N, 0)
print('word_features', word_features)
# 使用原始单词计数来作为度量指标
def doc_features(doc):
doc_words = FreqDist(w for w in doc if not isStopWord(w))
features = {}
for word in word_features:
features['count (%s)' % word] = (doc_words.get(word, 0))
return features
# 使用原始单词计数,来作为特征值
featuresets = [(doc_features(d), c) for (d, c) in labeled_docs]
# 创建训练数据集和测试数据集
train_set, test_set = featuresets[200:], featuresets[:200]
# 检查分类器效果
classifier = NaiveBayesClassifier.train(train_set)
# 根据测试数据集来计算分类器的准确性
print("Accuracy", accuracy(classifier, test_set))
# 贡献度最大的特征
print(classifier.show_most_informative_features())
运行结果如下:
#Review Words 1583820
# After filter 710579
Accuracy 0.765
Most Informative Features
count (wonderful) = 2 pos : neg = 14.8 : 1.0
count (outstanding) = 1 pos : neg = 12.0 : 1.0
count (apparently) = 2 neg : pos = 12.0 : 1.0
count (stupid) = 2 neg : pos = 11.1 : 1.0
count (boring) = 2 neg : pos = 10.7 : 1.0
count (bad) = 5 neg : pos = 10.0 : 1.0
count (best) = 4 pos : neg = 9.9 : 1.0
count (anyway) = 2 neg : pos = 8.1 : 1.0
count (minute) = 2 neg : pos = 8.1 : 1.0
count (matt) = 2 pos : neg = 7.9 : 1.0
None
7 创建词云
示例代码如下:
def getElementsByNum(values, num, keyOrValue):
"""
取得指定列表的num个元素的key或是value,
:param values:列表
:param num: 元素个数
:param keyOrValue: 元素的键还是值 0表示键,1表示值
:return:
"""
tmpValue = []
for key in sorted(values.items(), key=lambda d: d[1], reverse=True)[:num]:
tmpValue.append(key[keyOrValue])
return tmpValue
# 停用词和标点符号集合
sw = set(stopwords.words('english'))
punctuation = set(string.punctuation)
# 检查是否为停用词或是标点符号
def isStopWord(word):
return word in sw and word in punctuation
# 取得原始文档
review_words = movie_reviews.words()
# 过滤停用词和标点符号
filtered = [w.lower() for w in review_words if not isStopWord(w.lower())]
# 选用词频最高的前1%作为特征
words = FreqDist(filtered)
N = int(.01 * len(words.keys()))
tags = getElementsByNum(words, N, 0)
# tags = words.keys()[:N]
for tag in tags:
print(tag, ":", words[tag])
# 将输出结果粘粘到wordle页面,就可以得到词云页面
输出结果:略
进一步的过滤
词频和逆文档频率 The Term Frequency
示例代码如下:
from nltk.corpus import movie_reviews
from nltk.corpus import stopwords
from nltk.corpus import names
from nltk import FreqDist
from sklearn.feature_extraction.text import TfidfVectorizer
import itertools
import pandas as pd
import numpy as np
import string
# 设置停用词 标点符号和姓名
sw = set(stopwords.words('english'))
punctuation = set(string.punctuation)
all_names = set([name.lower() for name in names.words()])
# 过滤单词(停用词,标点符号,姓名,数字)
def isStopWord(word):
return (word in sw or word in punctuation) or not word.isalpha() or word in all_names
# 取得影评文档
review_words = movie_reviews.words()
# 过滤停用词
filtered = [w.lower() for w in review_words if not isStopWord(w.lower())]
words = FreqDist(filtered)
# 创建TfidfVectorizer所需要的字符串列表(过滤掉停用词和只出现一次的单词 )
texts = []
for fid in movie_reviews.fileids():
texts.append(" ".join([w.lower()
for w in movie_reviews.words(fid)
if not isStopWord(w.lower()) and words[w.lower()] > 1]))
# 创建向量化程序
vectorizer = TfidfVectorizer(stop_words='english')
matrix = vectorizer.fit_transform(texts)
# 求单词的TF-IDF的和
sums = np.array(matrix.sum(axis=0)).ravel()
# 通过单词的排名权值
ranks = []
# 不可用
# for word, val in itertools.izip(vectorizer.get_feature_names(), sums):
for word, val in zip(vectorizer.get_feature_names(), sums):
ranks.append((word, val))
# 创建DataFrame
df = pd.DataFrame(ranks, columns=['term', 'tfidf'])
# 并排序
# df = df.sort(columns='tfidf')
df = df.sort_values(by='tfidf')
# 输出排名字低的值
print(df.head())
N = int(.01 * len(df))
df = df.tail(N)
# 不可用
# for term, tfidf in itertools.izip(df['term'].values, df['tfidf'].values):
for term, tfidf in zip(df['term'].values, df['tfidf'].values):
print(term, ":", tfidf)
运行结果如下:
term tfidf
19963 superintendent 0.03035
8736 greys 0.03035
14010 ology 0.03035
2406 briefer 0.03035
2791 cannibalize 0.03035
matter : 10.1601563202
review : 10.1621092081
...
jokes : 10.1950553877
8 社交网络分析
安装networdX 略
利用网络理论来研究社会关系
示例代码如下:
import matplotlib.pyplot as plt
import networkx as nx
# NetwordX所的供的示例图
print([s for s in dir(nx) if s.endswith("graph")])
G = nx.davis_southern_women_graph()
plt.figure(1)
plt.hist(list(nx.degree(G).values()))
plt.figure(2)
pos = nx.spring_layout(G)
nx.draw(G, node_size=9)
nx.draw_networkx_labels(G, pos)
plt.show()
运行结果如下:
['LCF_graph', 'barabasi_albert_graph','barbell_graph', 'binomial_graph', 'bull_graph', 'caveman_graph','chordal_cycle_graph', 'chvatal_graph', 'circulant_graph','circular_ladder_graph', 'complete_bipartite_graph', 'complete_graph','complete_multipartite_graph', 'connected_caveman_graph','connected_watts_strogatz_graph', 'cubical_graph', 'cycle_graph','davis_southern_women_graph', 'dense_gnm_random_graph', 'desargues_graph','diamond_graph', 'digraph', 'directed_havel_hakimi_graph','dodecahedral_graph', 'dorogovtsev_goltsev_mendes_graph','duplication_divergence_graph', 'ego_graph', 'empty_graph','erdos_renyi_graph', 'expected_degree_graph', 'fast_gnp_random_graph','florentine_families_graph', 'frucht_graph', 'gaussian_random_partition_graph','general_random_intersection_graph', 'geographical_threshold_graph','gn_graph', 'gnc_graph', 'gnm_random_graph', 'gnp_random_graph', 'gnr_graph','graph', 'grid_2d_graph', 'grid_graph', 'havel_hakimi_graph', 'heawood_graph','house_graph', 'house_x_graph', 'hypercube_graph', 'icosahedral_graph','is_directed_acyclic_graph', 'k_random_intersection_graph','karate_club_graph', 'kl_connected_subgraph', 'krackhardt_kite_graph','ladder_graph', 'line_graph', 'lollipop_graph', 'make_max_clique_graph','make_small_graph', 'margulis_gabber_galil_graph', 'moebius_kantor_graph','multidigraph', 'multigraph', 'navigable_small_world_graph','newman_watts_strogatz_graph', 'null_graph', 'nx_agraph', 'octahedral_graph','pappus_graph', 'path_graph', 'petersen_graph', 'planted_partition_graph','powerlaw_cluster_graph', 'projected_graph', 'quotient_graph','random_clustered_graph', 'random_degree_sequence_graph','random_geometric_graph', 'random_partition_graph', 'random_regular_graph','random_shell_graph', 'relabel_gexf_graph', 'relaxed_caveman_graph','scale_free_graph', 'sedgewick_maze_graph', 'star_graph', 'stochastic_graph','subgraph', 'tetrahedral_graph', 'to_networkx_graph', 'trivial_graph','truncated_cube_graph', 'truncated_tetrahedron_graph', 'tutte_graph','uniform_random_intersection_graph', 'watts_strogatz_graph', 'waxman_graph','wheel_graph']