NLP之word2vec：利用 Wikipedia Text(中文維基百科)語(yǔ)料+Word2vec工具來(lái)訓練簡(jiǎn)體中文詞向量
后期更新……
最后的model
word2vec_wiki.model.rar
后期更新……
Wikipedia Text語(yǔ)料來(lái)源及其下載：zhwiki dump progress on 20190120
? ? ? ?其中zhwiki-latest-pages-articles.xml.bz2文件包含了標題、正文部分。壓縮包大概是1.3G，解壓后大概是5.7G。相比英文wiki中文的還是小了不少。
? ? ? ? 下載下來(lái)的wiki是XML格式，需要提取其正文內容。不過(guò)維基百科的文檔解析有不少的成熟工具（例如gensim，wikipedia extractor等。其中Wikipedia Extractor?是一個(gè)簡(jiǎn)單方便的Python腳本。
T1、Wikipedia Extractor工具
Wikipedia extractor的網(wǎng)址：?http://medialab.di.unipi.it/wiki/Wikipedia_Extractor
Wikipedia extractor的使用：下載好WikiExtractor.py后直接使用下面的命令運行即可，
?? ? ? ? ? ? ? ? ??其中，-cb 1200M表示以 1200M 為單位切分文件，-o 后面接出入文件，最后是輸入文件。
WikiExtractor.py -cb 1200M -o extracted zhwiki-latest-pages-articles.xml.bz2T2、python代碼實(shí)現
?? ?將這個(gè)XML壓縮文件轉換為txt文件
python process_wiki.py zhwiki-latest-pages-articles.xml.bz2 wiki.zh.text?? ? ?中文wiki內容中大多數是繁體，這需要進(jìn)行簡(jiǎn)繁轉換?？梢圆捎脧B門(mén)大學(xué)NLP實(shí)驗室開(kāi)發(fā)的簡(jiǎn)繁轉換工具或者opencc代碼實(shí)現。
T1、廈門(mén)大學(xué)NLP實(shí)驗室開(kāi)發(fā)的簡(jiǎn)繁轉換工具
轉換工具下載網(wǎng)址：http://jf.cloudtranslation.cc/
轉換工具的使用：下載單機版即可，在windos命令行窗口下使用下面命令行運行
?? ? ? ? ? ? ? ? ??其中file1.txt為繁體原文文件，file2.txt為輸出轉換結果的目標文件名，lm_s2t.txt為語(yǔ)言模型文件。
jf -fj file1.txt file2.txt -lm lm_s2t.txtT2、opencc代碼實(shí)現
opencc -i wiki.zh.text -o wiki.zh.text.jian -c zht2zhs.ini, 將繁體字轉換為簡(jiǎn)體字。iconv -c -t UTF-8 < wiki.zh.text.jian.seg > wiki.zh.text.jian.seg.utf-8python train_word2vec_model.py wiki.zh.text.jian.seg.utf-8 wiki.zh.text.model wiki.zh.text.vector正在更新……
對下邊文件代碼的說(shuō)明
#We create word2vec model use wiki Text like this https://dumps.wikimedia.org/zhwiki/20161001/zhwiki-20161001-pages-articles-multistream.xml.bz2
##parameter:
=================================
? ? feature_size = 500
? ? content_window = 5
? ? freq_min_count = 3
? ? # threads_num = 4
? ? negative = 3 ? #best采樣使用hierarchical softmax方法(負采樣，對常見(jiàn)詞有利)，不使用negative sampling方法(對罕見(jiàn)詞有利)。
? ? iter = 20
##process.py deal with wiki*.xml
##word2vec_wiki.py : create model and load model1、process.py文件
#process.py文件
import logging
import os.path
import sys
from gensim.corpora import WikiCorpus
#run python process_wiki.py ../data/zhwiki-latest-pages-articles.xml.bz2 wiki.zh.text
if __name__ == '__main__':
program = os.path.basename(sys.argv[0])
logger = logging.getLogger(program)
logging.basicConfig(format='%(asctime)s: %(levelname)s: %(message)s')
logging.root.setLevel(level=logging.INFO)
logger.info("running %s" % ' '.join(sys.argv))
# check and process input arguments
if len(sys.argv) < 3:
print(('globals()[__doc__])：', globals()['__doc__']) )
print('locals：', locals())
print(globals()['__doc__'] % locals()) #最初代碼 print globals()['__doc__'] % locals()
sys.exit(1)
inp, outp = sys.argv[1:3]
space = " "
i = 0
output = open(outp, 'w')
wiki = WikiCorpus(inp, lemmatize=False, dictionary={})
for text in wiki.get_texts():
output.write(space.join(text) + "\n")
i = i + 1
if (i % 10000 == 0):
logger.info("Saved " + str(i) + " articles")
output.close()
logger.info("Finished Saved " + str(i) + " articles")2、word2vec_wiki.py文件
#word2vec_wiki.py文件
# -*- coding:utf-8 -*-
from __future__ import print_function
import numpy as np
import os
import sys
import jieba
import time
import jieba.posseg as pseg
import codecs
import multiprocessing
import json
# from gensim.models import Word2Vec,Phrases
from gensim import models,corpora
import logging
# auto_brand = codecs.open("Automotive_Brand.txt", encoding='utf-8').read()
sys.path.append("../../")
sys.path.append("../../langconv/")
sys.path.append("../../parser/")
# import xmlparser
# from xmlparser import *
# from langconv import *
# logger = logging.getLogger(program)
logging.basicConfig(format='%(asctime)s: %(levelname)s: %(message)s')
logging.root.setLevel(level=logging.INFO)
# logger.info("running %s" % ' '.join(sys.argv))
def json_dict_from_file(json_file,fieldnames=None,isdelwords=True):
"""
load json file and generate a new object instance whose __name__ filed
will be 'inst'
:param json_file:
"""
obj_s = []
with open(json_file) as f:
for line in f:
object_dict = json.loads(line)
if fieldnames==None:
obj_s.append(object_dict)
else:
# for fieldname in fieldname:
if set(fieldnames).issubset(set(object_dict.keys())):
one = []
for fieldname in fieldnames:
if isdelwords and fieldname == 'content':
one.append(delNOTNeedWords(object_dict[fieldname])[1])
else:
one.append(object_dict[fieldname])
obj_s.append(one)
return obj_s
def delNOTNeedWords(content,customstopwords=None):
# words = jieba.lcut(content)
if customstopwords == None:
customstopwords = "stopwords.txt"
import os
if os.path.exists(customstopwords):
stop_words = codecs.open(customstopwords, encoding='UTF-8').read().split(u'\n')
customstopwords = stop_words
result=''
return_words = []
# for w in words:
# if w not in stopwords:
# result += w.encode('utf-8') # +"/"+str(w.flag)+" " #去停用詞
words = pseg.lcut(content)
for word, flag in words:
# print word.encode('utf-8')
tempword = word.encode('utf-8').strip(' ')
if (word not in customstopwords and len(tempword)>0 and flag in [u'n',u'nr',u'ns',u'nt',u'nz',u'ng',u't',u'tg',u'f',u'v',u'vd',u'vn',u'vf',u'vx',u'vi',u'vl',u'vg', u'a',u'an',u'ag',u'al',u'm',u'mq',u'o',u'x']):
# and flag[0] in [u'n', u'f', u'a', u'z']):
# ["/x","/zg","/uj","/ul","/e","/d","/uz","/y"]): #去停用詞和其他詞性，比如非名詞動(dòng)詞等
result += tempword # +"/"+str(w.flag)+" " #去停用詞
return_words.append(tempword)
return result,return_words
def get_save_wikitext(wiki_filename,text_filename):
output = open(text_filename, 'w')
wiki = corpora.WikiCorpus(text_filename, lemmatize=False, dictionary={})
for text in wiki.get_texts():
# text = delNOTNeedWords(text,"../../stopwords.txt")[1]
output.write(" ".join(text) + "\n")
i = i + 1
if (i % 10000 == 0):
logging.info("Saved " + str(i) + " articles")
output.close()
def load_save_word2vec_model(line_words, model_filename):
# 模型參數
feature_size = 500
content_window = 5
freq_min_count = 3
# threads_num = 4
negative = 3 #best采樣使用hierarchical softmax方法(負采樣，對常見(jiàn)詞有利)，不使用negative sampling方法(對罕見(jiàn)詞有利)。
iter = 20
print("word2vec...")
tic = time.time()
if os.path.isfile(model_filename):
model = models.Word2Vec.load(model_filename)
print(model.vocab)
print("Loaded word2vec model")
else:
bigram_transformer = models.Phrases(line_words)
model = models.Word2Vec(bigram_transformer[line_words], size=feature_size, window=content_window, iter=iter, min_count=freq_min_count,negative=negative, workers=multiprocessing.cpu_count())
toc = time.time()
print("Word2vec completed! Elapsed time is %s." % (toc-tic))
model.save(model_filename)
# model.save_word2vec_format(save_model2, binary=False)
print("Word2vec Saved!")
return model
if __name__ == '__main__':
limit = -1 #該屬性決定取wiki文件text tag前多少條，-1為所有
wiki_filename = '/home/wac/data/zhwiki-20160203-pages-articles-multistream.xml'
wiki_text = './wiki_text.txt'
wikimodel_filename = './word2vec_wiki.model'
s_list = []
# if you want get wiki text ,uncomment lines
# get_save_wikitext(wiki_filename,wiki_text)
# for i,text in enumerate(open(wiki_text, 'r')):
# s_list.append(delNOTNeedWords(text,"../../stopwords.txt")[1])
# print(i)
#
# if i==limit: #取前l(fā)imit條,-1為所有
# break
#
#計算模型
model = load_save_word2vec_model(s_list, wikimodel_filename)
#計算相似單詞，命令行輸入
while 1:
print("請輸入想測試的單詞： ", end='\b')
t_word = sys.stdin.readline()
if "quit" in t_word:
break
try:
results = model.most_similar(t_word.decode('utf-8').strip('\n').strip('\r').strip(' ').split(' '), topn=30)
except:
continue
for t_w, t_sim in results:
print(t_w, " ", t_sim)
參考文章(貼上源址表示感謝)
使用維基百科訓練簡(jiǎn)體中文詞向量
中文Wiki語(yǔ)料獲取
Wiki語(yǔ)料處理
中文維基語(yǔ)料訓練獲取
Windows3.5下對維基百科語(yǔ)料用word2vec進(jìn)行訓練尋找同義詞相似度
聯(lián)系客服
