搜狗新闻语料处理训练词向量

首先，去https://www.sogou.com/labs/resource/ca.php下载语料，我下的是完整版711M数据。解压生成dat文件，并不是utf-8格式先要转换：

cat news_tensite_xml.dat | iconv -f gbk -t utf-8 -c | grep "<content>" > news.txt'

文件较大，后续处理可能不方便，所以将大语料切分到split_files目录下：

#! /usr/bin/env Python3
# -*- coding:utf8 -*- 
import re       #正则表达式

def split_file(base_dir, input_file, max_lines):
    f = open(base_dir + input_file, "r")
    line_index = 0 #每个txt行数
    txt_num = 0 #txt数目
    output = open(base_dir + "split_files/" + str(txt_num) + ".txt", 'w+')
    for line in f:
     
        if "</content>" in line and line_index > max_lines: #一篇文章结束且达到此t x t最大行
            #写入此t x t最后一行，并开一篇新t x t，行数重设为0
            output.write(line + "\n")
            output.close()
            txt_num += 1
            output = open(base_dir + "split_files/" + str(txt_num) + ".txt", 'w+')
            line_index = 0
            
            print(txt_num)
            
            
        else:
            #写入此txt
            output.write(line + "\n")
            line_index += 1

    
split_file("data/sougou_corpus/", "news.txt", 100000)

然后提取content标签内容，分词后放在cut目录下。

import codecs
import sys
import re
import csv
import os
import codecs

rootdir = 'data/sougou_corpus/split_files'
list = os.listdir(rootdir)
pattern1 = r'<content>(.*?)</content>'


for x in range(0,len(list)):
   path = os.path.join(rootdir, list[x])        #获取目录下文件名字
  
   if os.path.isfile(path) and list[x].endswith(".txt"):
      print(path)
      output = open("data/sougou_corpus/cut/" + list[x], "w+", encoding = "utf-8")      
      content = open(path,'r').read()
      texts = re.findall(pattern1,content,re.S)
      for text in texts:
         if text:     
            line = ' '.join(jieba.lcut(text))
            output.write(line + "\n")
      output.close()

用一个文件训练词向量并保存：

#coding=utf-8
from gensim.models import word2vec

sentences = word2vec.Text8Corpus("data/sougou_corpus/cut/12.txt")  # 加载语料
model = word2vec.Word2Vec(sentences, size=200)  # 默认window=5
model.save('data/sougou_corpus/models/word2vec.model')

查看效果：
image.png
还不错吧。

还可以继续训练其他文本。

def word2vec_train(input_file):

    model = word2vec.Word2Vec.load("models/word2vec.model")
    sentences = word2vec.Text8Corpus(input_file)  # 加载语料
    model.train(sentences, epochs=model.iter, total_examples=model.corpus_count)
    model.save('models/word2vec.model')