nltk-コーパスの構築と使用-小説に使用できる推奨-完全なインスタンス
手順1:コーパスの構築:
データソース:
からhttp://d1.txthj.com/newrar/txthj_264.rarの83編の小説をカタログに保存する./data/下.
ロード時に二層ディレクトリとして処理
出力:
./dictおよび./corpus
対応ディレクトリの下でxxxを生成する.dictとxxx.mm,xxxは元のファイルのフルネーム(パスを含まない、接尾辞を含む)
手順2:コーパスのロード、類似性分析
説明:
yieldの使用は、より良いメモリ効率のためです.
残された問題:
手順2でヒントが表示されます.
/usr/lib/python2.7/dist-packages/scipy/sparse/compressed.py:122: UserWarning: indices array has non-integer dtype (float64)
プロセスに影響しない
#!/usr/bin/env python
#-*-coding=utf-8-*-
# ( )
sourceDataDir='data'
#
fileLists = []
import os
from gensim import corpora, models, similarities
def getSourceFileLists(sourceDataDir):
fileLists = []
subDirList = os.listdir(sourceDataDir)
for subDir in subDirList:
subList = os.listdir(sourceDataDir + '/' + subDir)
fileList = [ sourceDataDir+'/'+subDir+'/'+ x for x in subList if os.path.isfile(sourceDataDir+'/'+subDir+'/'+x)]
fileLists += fileList
return fileLists
fileLists = getSourceFileLists(sourceDataDir)
if 0 < len(fileLists):
import codecs
import jieba
punctuations = ['','
','\t',',', '.', ':', ';', '?', '(', ')', '[', ']', '&', '!', '*', '@', '#', '$', '%']
if not os.path.exists('dict'):
os.mkdir("dict")
if not os.path.exists('corpus'):
os.mkdir("corpus")
for fileName in fileLists:
print fileName
hFile = None
content = None
try:
hFile = codecs.open(fileName,'r','gb18030')
content = hFile.readlines()
except Exception,e:
print e
finally:
if hFile:
hFile.close()
if content:
fileFenci = [ x for x in jieba.cut(' '.join(content),cut_all=True)]
fileFenci2 = [word for word in fileFenci if not word in punctuations]
texts = [fileFenci2]
all_tokens = sum(texts, [])
tokens_once = set(word for word in set(all_tokens) if all_tokens.count(word) == 1)
texts = [[word for word in text if word not in tokens_once] for text in texts]
sFileDir, sFileName = os.path.split(fileName)
dictFileName = 'dict/'+sFileName+'.dict'
corpusFileName = 'corpus/'+sFileName+'.mm'
dictionary = corpora.Dictionary(texts)
dictionary.save_as_text(dictFileName)
corpus = ([dictionary.doc2bow(text) for text in texts])
corpora.MmCorpus.serialize(corpusFileName, corpus)
print 'Build corpus done'データソース:
からhttp://d1.txthj.com/newrar/txthj_264.rarの83編の小説をカタログに保存する./data/下.
ロード時に二層ディレクトリとして処理
出力:
./dictおよび./corpus
対応ディレクトリの下でxxxを生成する.dictとxxx.mm,xxxは元のファイルのフルネーム(パスを含まない、接尾辞を含む)
手順2:コーパスのロード、類似性分析
#!/usr/bin/env python
#-*-coding=utf-8-*-
import os
from gensim import corpora, models, similarities
def getFileList(dir):
return [ dir + x for x in os.listdir(dir)]
dictLists = getFileList('./dict/')
class LoadDictionary(object):
def __init__(self, dictionary):
self.dictionary = dictionary
def __iter__(self):
for dictFile in dictLists:
sFileRaw, sFilePostfix = os.path.splitext(dictFile)
sFileDir, sFileName = os.path.split(sFileRaw)
(dictFile, corpusFile) = ( './dict/' + sFileName + '.dict', './corpus/'+sFileName + '.mm')
yield self.dictionary.load_from_text(dictFile)
class LoadCorpus(object):
def __iter__(self):
for dictFile in dictLists:
sFileRaw, sFilePostfix = os.path.splitext(dictFile)
sFileDir, sFileName = os.path.split(sFileRaw)
(dictFile, corpusFile) = ( './dict/' + sFileName + '.dict', './corpus/'+sFileName + '.mm')
yield corpora.MmCorpus(corpusFile)
"""
(easy_install nltk)
"""
# +
def pre_process_cn(inputs, low_freq_filter = True):
"""
1.
2.
3.
4.
"""
import nltk
import jieba.analyse
from nltk.tokenize import word_tokenize
texts_tokenized = []
for document in inputs:
texts_tokenized_tmp = []
for word in word_tokenize(document):
texts_tokenized_tmp += jieba.analyse.extract_tags(word,10)
texts_tokenized.append(texts_tokenized_tmp)
texts_filtered_stopwords = texts_tokenized
#
english_punctuations = [',', '.', ':', ';', '?', '(', ')', '[', ']', '&', '!', '*', '@', '#', '$', '%']
texts_filtered = [[word for word in document if not word in english_punctuations] for document in texts_filtered_stopwords]
#
from nltk.stem.lancaster import LancasterStemmer
st = LancasterStemmer()
texts_stemmed = [[st.stem(word) for word in docment] for docment in texts_filtered]
#
if low_freq_filter:
all_stems = sum(texts_stemmed, [])
stems_once = set(stem for stem in set(all_stems) if all_stems.count(stem) == 1)
texts = [[stem for stem in text if stem not in stems_once] for text in texts_stemmed]
else:
texts = texts_stemmed
return texts
dictionary = corpora.dictionary.Dictionary()
dictionary_memory_friendly = LoadDictionary(dictionary)
for vector in dictionary_memory_friendly:
dictionary = vector
corpus = []
corpus_memory_friendly = LoadCorpus()
for vector in corpus_memory_friendly:
corpus.append(vector[0])
if 0 < len(corpus):
tfidf = models.TfidfModel(corpus)
corpus_tfidf = tfidf[corpus]
model = models.LsiModel(corpus_tfidf, id2word=None, num_topics=20, chunksize=2000000) # id2word=dictionary ，LsiModel corpus dictionary
index = similarities.Similarity('./novel_', model[corpus], num_features=len(corpus))
# ，
target_courses = [' ， ， ， 。']
target_text = pre_process_cn(target_courses, low_freq_filter=False)
"""
"""
#
ml_course = target_text[0]
#
ml_bow = dictionary.doc2bow(ml_course)
# lsi model ，
ml_lsi = model[ml_bow] #ml_lsi (topic_id, topic_value)
sims = index[ml_lsi] #sims ， index[xxx] __getitem__() ml_lsi
# ，
sort_sims = sorted(enumerate(sims), key=lambda item: -item[1])
#
print sort_sims[0:10]
print len(dictLists)
print dictLists[sort_sims[1][0]]
print dictLists[sort_sims[2][0]]
print dictLists[sort_sims[3][0]]説明:
yieldの使用は、より良いメモリ効率のためです.
残された問題:
手順2でヒントが表示されます.
/usr/lib/python2.7/dist-packages/scipy/sparse/compressed.py:122: UserWarning: indices array has non-integer dtype (float64)
プロセスに影響しない