NLP词向量和句向量方法总结及实现

from gensim.models.word2vec import Word2Vec
import pandas as pd

# 构建word2vec模型，词向量的训练与生成
def get_dataset_vec(dataset):
    n_dim = 300
    w2v_model = Word2Vec(dataset, sg=1, size=n_dim, min_count=10, hs=0)  # 初始化模型并训练
    # 在测试集上训练
    # w2v_model.train(x_test,total_examples=w2v_model.corpus_count,epochs=w2v_model.iter) #追加训练模型
    # 将imdb_w2v模型保存，训练集向量，测试集向量保存到文件
    # print(w2v_model['会议'])
    w2v_model.save('data/w2v/w2v_model_300.pkl')  # 保存训练结果

if __name__ == '__main__':
    # 数据集获取
    train_data = pd.read_csv('data/clean_data_train.csv', sep=',', names=['contents', 'labels']).astype(str)
    test_data = pd.read_csv('data/clean_data_test.csv', sep=',', names=['contents', 'labels']).astype(str)
    cw = lambda x: str(x).split()
    train_data['words'] = train_data['contents'].apply(cw)
    test_data['words'] = train_data['contents'].apply(cw)
    dataset = pd.concat([train_data, test_data])

    # word2vec词向量训练
    get_dataset_vec(dataset['words'])

    # 词向量模型加载
    # w2v_model = Word2Vec.load('data/w2v/w2v_model_300.pkl')

word2vec 相关的API都在包gensim.models.word2vec中　，主要参数如下：　　　

sentences: 我们要分析的语料，可以是一个列表，或者从文件中遍历读出。后面我们会有从文件读出的例子。
size: 词向量的维度，默认值是100。这个维度的取值一般与我们的语料的大小相关，如果是不大的语料，比如小于100M的文本语料，则使用默认值一般就可以了。如果是超大的语料，建议增大维度。
window：即词向量上下文最大距离，这个参数在我们的算法原理篇中标记为cc，window越大，则和某一词较远的词也会产生上下文关系。默认值为5。在实际使用中，可以根据实际的需求来动态调整这个window的大小。如果是小语料则这个值可以设的更小。对于一般的语料这个值推荐在[5,10]之间。
sg: 即我们的word2vec两个模型的选择了。如果是0，则是CBOW模型，是1则是Skip-Gram模型，默认是0即CBOW模型。
hs: 即我们的word2vec两个解法的选择了，如果是0，则是Negative Sampling，是1的话并且负采样个数negative大于0，则是Hierarchical Softmax。默认是0即Negative Sampling。
negative:即使用Negative Sampling时负采样的个数，默认是5。推荐在[3,10]之间。这个参数在我们的算法原理篇中标记为neg。
cbow_mean: 仅用于CBOW在做投影的时候，为0，则算法中的xwxw为上下文的词向量之和，为1则为上下文的词向量的平均值。在我们的原理篇中，是按照词向量的平均值来描述的。个人比较喜欢用平均值来表示xwxw,默认值也是1,不推荐修改默认值。
min_count:需要计算词向量的最小词频。这个值可以去掉一些很生僻的低频词，默认是5。如果是小语料，可以调低这个值。
iter: 随机梯度下降法中迭代的最大次数，默认是5。对于大语料，可以增大这个值。
alpha: 在随机梯度下降法中迭代的初始步长。算法原理篇中标记为ηη，默认是0.025。
min_alpha: 由于算法支持在迭代的过程中逐渐减小步长，min_alpha给出了最小的迭代步长值。随机梯度下降中每轮的迭代步长可以由iter，alpha， min_alpha一起得出。这部分由于不是word2vec算法的核心内容，因此在原理篇我们没有提到。对于大语料，需要对alpha, min_alpha,iter一起调参，来选择合适的三个值。

3、基于Word2Vec的句向量

计算公式如下：

实现代码如下：


#对每个句子的所有词向量取均值，来生成一个句子的vector
def build_sentence_vector(sentence,size,w2v_model):
    sen_vec=np.zeros(size).reshape((1,size))
    count=0
    for word in sentence:
        try:
            sen_vec+=w2v_model[word].reshape((1,size))
            count+=1
        except KeyError:
            continue
    if count!=0:
        sen_vec/=count
    return sen_vec

4、基于加权Word2Vec的句向量

计算公式如下：

实现代码如下：


#对每个句子的所有词向量取加权均值，来生成一个句子的vector
def build_sentence_vector_weight(sentence,size,w2v_model,key_weight):
    key_words_list=list(key_weight)
    sen_vec=np.zeros(size).reshape((1,size))
    count=0
    for word in sentence:
        try:
            if word in key_words_list:
                sen_vec+=(np.dot(w2v_model[word],math.exp(key_weight[word]))).reshape((1,size))
                count+=1
            else:
                sen_vec+=w2v_model[word].reshape((1,size))
                count+=1
        except KeyError:
            continue
    if count!=0:
        sen_vec/=count
    return sen_vec

5、基于Word2Vec的文本向量化实现


# 将文本数据转换为文本向量
def doc_vec():
    train_data = pd.read_csv('data/clean_data_train.csv', sep=',',names=['contents', 'labels']).astype(str)
    test_data = pd.read_csv('data/clean_data_test.csv', sep=',', names=['contents', 'labels']).astype(str)
    w2v_model = Word2Vec.load('data/w2v/w2v_model_300.pkl')   #加载训练好的Word2Vec模型
 
    #读取词权重字典
    #with open('data/key_words_importance', 'r') as f:
       #key_words_importance = eval(f.read())
 
    cw=lambda x:int(x)
    y_train = np.array(train_data['labels'].apply(cw))
    y_test=np.array(test_data['labels'].apply(cw))
 
    #训练集转换为向量
    train_lenth=len(train_data)
    train_data_list=[]
    for i in range(train_lenth):
        train_data_list.append(str(train_data['contents'][i]).split())
    train_docvec_list=np.concatenate([build_sentence_vector(sen,300,w2v_model) for sen in train_data_list])
 
    #测试集转换为向量
    test_lenth = len(test_data)
    test_data_list = []
    for i in range(test_lenth):
        test_data_list.append(str(test_data['contents'][i]).split())
    test_docvec_list = np.concatenate([build_sentence_vector(sen, 300, w2v_model) for sen in test_data_list])
 
    return train_docvec_list,y_train,test_docvec_list,y_test

二、GloVe

1、GloVe介绍

现有词向量模型中，第一种是全局的词-文本矩阵分解（LSA），该方法能有效收集每一个词的统计信息，但他们却不能捕捉到词的上下文信息（语义的表达能力不够）；第二种就是基于局部窗口信息（Word2Vec），这种方法虽然能在词的语义上有更丰富的表达，但是他们却不能很好的捕捉词的全局统计信息。

GloVe词向量模型融合了全局矩阵分解方法（Matrix Factorization）和局部文本框捕捉方法（word2vec），是一种用于获得单词矢量表示的无监督学习算法。

模型目标：进行词的向量化表示，使得向量之间尽可能多地蕴含语义和语法的信息。
输入：语料库
输出：词向量
方法概述：首先基于语料库构建词的共现矩阵，然后基于共现矩阵和GloVe模型学习词向量。

GloVe论文 GloVe源码 GloVe介绍斯坦福NLP课程

2、基于源码的GloVe词向量生成（Linux下实现）

首先，下载GloVe源码

命令：cd到GloVe源码对应目录下，vim demo.sh

demo.sh文件如下：

#!/bin/bash

# Makes programs, downloads sample data, trains a GloVe model, and then evaluates it.
# One optional argument can specify the language used for eval script: matlab, octave or [default] python

make
if [ ! -e text8 ]; then
  if hash wget 2>/dev/null; then
    wget http://mattmahoney.net/dc/text8.zip
  else
    curl -O http://mattmahoney.net/dc/text8.zip
  fi
  unzip text8.zip
  rm text8.zip
fi

CORPUS=text8
VOCAB_FILE=vocab.txt
COOCCURRENCE_FILE=cooccurrence.bin
COOCCURRENCE_SHUF_FILE=cooccurrence.shuf.bin
BUILDDIR=build
SAVE_FILE=vectors
VERBOSE=2
MEMORY=4.0
VOCAB_MIN_COUNT=5
VECTOR_SIZE=50
MAX_ITER=15
WINDOW_SIZE=15
BINARY=2
NUM_THREADS=8
X_MAX=10

$BUILDDIR/vocab_count -min-count $VOCAB_MIN_COUNT -verbose $VERBOSE < $CORPUS > $VOCAB_FILE
if [[ $? -eq 0 ]]
  then
  $BUILDDIR/cooccur -memory $MEMORY -vocab-file $VOCAB_FILE -verbose $VERBOSE -window-size $WINDOW_SIZE < $CORPUS > $COOCCURRENCE_FILE
  if [[ $? -eq 0 ]]
  then
    $BUILDDIR/shuffle -memory $MEMORY -verbose $VERBOSE < $COOCCURRENCE_FILE > $COOCCURRENCE_SHUF_FILE
    if [[ $? -eq 0 ]]
    then
       $BUILDDIR/glove -save-file $SAVE_FILE -threads $NUM_THREADS -input-file $COOCCURRENCE_SHUF_FILE -x-max $X_MAX -iter $MAX_ITER -vector-size $VECTOR_SIZE -binary $BINARY -vocab-file $VOCAB_FILE -verbose $VERBOSE
       if [[ $? -eq 0 ]]
       then
           if [ "$1" = 'matlab' ]; then
               matlab -nodisplay -nodesktop -nojvm -nosplash < ./eval/matlab/read_and_evaluate.m 1>&2 
           elif [ "$1" = 'octave' ]; then
               octave < ./eval/octave/read_and_evaluate_octave.m 1>&2 
           else
               python eval/python/evaluate.py
           fi
       fi
    fi
  fi
fi

对demo.sh文件进行修改

把make这边注释掉，这个是让你去下个demo，我们直接改成自己的数据
CORPUS需要对应自己的欲训练的文档
词向量维度，窗口大小，单词至少出现的次数，迭代次数等参数按需修改

修改后的demo.sh文件如下

#!/bin/bash

# Makes programs, downloads sample data, trains a GloVe model, and then evaluates it.
# One optional argument can specify the language used for eval script: matlab, octave or [default] python

#请把make这边注释掉，这个是让你去下个demo，我们直接改成自己的数据
#make
#if [ ! -e text8 ]; then
#  if hash wget 2>/dev/null; then
#    wget http://mattmahoney.net/dc/text8.zip
#  else
#    curl -O http://mattmahoney.net/dc/text8.zip
#  fi
#  unzip text8.zip
#  rm text8.zip
#fi

CORPUS=counts.txt    #CORPUS需要对应自己的欲训练的文档
VOCAB_FILE=vocab.txt
COOCCURRENCE_FILE=cooccurrence.bin
COOCCURRENCE_SHUF_FILE=cooccurrence.shuf.bin
BUILDDIR=build
SAVE_FILE=vectors
VERBOSE=2
MEMORY=4.0
VOCAB_MIN_COUNT=5   #单词至少出现的次数
VECTOR_SIZE=300     #训练的词向量维度
MAX_ITER=15         #训练迭代次数
WINDOW_SIZE=15      #窗口大小
BINARY=2
NUM_THREADS=8
X_MAX=10

$BUILDDIR/vocab_count -min-count $VOCAB_MIN_COUNT -verbose $VERBOSE < $CORPUS > $VOCAB_FILE
if [[ $? -eq 0 ]]
  then
  $BUILDDIR/cooccur -memory $MEMORY -vocab-file $VOCAB_FILE -verbose $VERBOSE -window-size $WINDOW_SIZE < $CORPUS > $COOCCURRENCE_FILE
  if [[ $? -eq 0 ]]
  then
    $BUILDDIR/shuffle -memory $MEMORY -verbose $VERBOSE < $COOCCURRENCE_FILE > $COOCCURRENCE_SHUF_FILE
    if [[ $? -eq 0 ]]
    then
       $BUILDDIR/glove -save-file $SAVE_FILE -threads $NUM_THREADS -input-file $COOCCURRENCE_SHUF_FILE -x-max $X_MAX -iter $MAX_ITER -vector-size $VECTOR_SIZE -binary $BINARY -vocab-file $VOCAB_FILE -verbose $VERBOSE
       if [[ $? -eq 0 ]]
       then
           if [ "$1" = 'matlab' ]; then
               matlab -nodisplay -nodesktop -nojvm -nosplash < ./eval/matlab/read_and_evaluate.m 1>&2 
           elif [ "$1" = 'octave' ]; then
               octave < ./eval/octave/read_and_evaluate_octave.m 1>&2 
           else
               python eval/python/evaluate.py
           fi
       fi
    fi
  fi
fi

训练文档为counts.txt，部分内容如下：

紧急通知 通知 七点 50 准时 武陵 楼 开会 早 开 早 散 十一 十一点 十一点半 一点 一点半 准时 散会 时间 会议
党 服 成员 明天 中午 12 点 30 分 第一 第一次 例会 会地 地点 待定 请 预留 留出 时间 收到 请 回复 全体成员 成员 时间 会议
通知 宣传 宣传部 部将 明天 中午 1230 周三 j4101 开本 本次 部门 例会 请 带好 笔 本次 例会 考评 部 本次 例会 主席 主席团 发 邀请 梁 淑 楠 收到 请 回复 全体成员 成员 时间 会议 议会 会议 议会 会议
本周 周六 叉叉 裤 聚会 聚会时间 时间 会议
总 通知 当班 班长 眼 长 表工 930 中控室 开会 开会时间 时间 会议

命令：sh demo.sh ，会得到vectors.txt，这个里面就对应每个词的向量表示。

vectors.txt部分内容如下：

时间 -0.192653 -0.027064 0.157989 -0.959652 -0.249349 -0.695960 -0.290393 -0.320663 0.850561 1.123161 -0.820032 -0.031310 -0.220525 0.391944 0.363133 0.433053 0.133621 0.590930 -1.096903 -0.346236 0.211173 1.074069 1.840458 -0.619614 -0.047013 -0.482248 -0.980474 -0.052697 -0.764487 -1.042479 0.310801 1.027933 -0.703576 -0.121927 0.096973 -0.389311 0.467025 0.532758 -0.626837 0.689972 0.020686 0.654299 -0.463879 0.430281 -0.206859 -0.253277 -0.091177 -0.616631 0.082589 -0.110885 0.858724 1.039171 -0.215845 0.597427 -0.564434 1.393491 -0.340165 0.079653 -0.093698 0.238416 0.800182 0.019553 -0.275385 0.314701 0.126838 -0.891766 -0.598391 -0.339991 0.278942 0.150008 0.169043 -0.106491 0.129413 0.154625 -1.077015 0.636639 0.760766 0.291428 -0.304663 -0.709912 0.578007 0.400317 0.498921 0.299688 -1.924125 -0.233359 -0.593315 -0.064116 0.141521 0.708793 0.896055 -0.439493 0.040833 0.089314 -0.004766 0.501943 0.239141 -0.368876 0.016068 -0.074727 -0.526003 -0.683604 1.052587 -0.511065 -1.180889 -0.136742 -0.571856 0.072116 -0.220324 -0.180080 0.959551 0.180546 0.292177 0.048728 -0.535014 0.014530 0.631147 -0.088323 0.668570 -0.310500 0.363473 -0.564667 0.857510 -0.148112 0.238181 0.135603 -0.073686 0.755832 0.572558 0.281101 0.297423 -0.748391 -0.020244 -0.557294 -0.441473 0.106799 -0.297012 0.233237 0.218440 -0.034616 0.410545 -0.245199 0.695645 0.192339 -1.024643 -0.628238 0.357365 0.426126 0.290365 -0.421078 -0.207671 -0.393413 0.527472 -0.465794 0.134838 -0.118478 -0.358524 0.733155 0.181742 -0.357532 -0.425675 0.262033 0.192739 0.320736 -0.073291 -0.603243 -0.214388 0.251359 -0.157907 0.639871 -0.322949 0.153679 0.128943 -0.079536 -1.243106 0.405427 -0.679834 0.706061 -0.556909 0.317398 -0.972522 -0.323571 0.082381 0.952649 0.470793 -0.255887 -0.072567 0.154703 0.553813 0.444109 -0.955931 -0.506380 -0.185121 0.436316 -0.166778 -0.070471 0.134493 -0.041048 0.371387 -0.466888 -0.322176 -0.139643 -0.469103 -0.297749 -0.341331 -0.182869 -1.138910 -0.840458 -1.005790 -1.030163 0.435325 -0.218241 -1.017183 -0.249351 0.060505 0.819773 -1.382254 -0.520934 -0.643470 0.369318 0.390774 -0.547370 -0.003034 -0.119246 0.397086 -0.720554 -0.267369 -0.626506 0.650304 0.491474 0.097891 0.035371 0.471716 0.680434 -0.168340 -0.088465 0.200047 -0.021552 -0.849985 0.090245 1.103723 0.242453 -0.830361 -0.132980 0.391840 -0.074169 -0.254151 0.175168 -0.482801 0.787707 1.463750 -0.666707 -0.324269 0.170228 -0.455849 -0.454797 0.485042 0.906702 -0.069496 -0.017353 0.824049 0.567829 -0.075913 -0.094140 -0.010110 -0.428374 0.830554 -0.249833 0.168137 0.541595 -1.243334 0.625043 -0.149822 0.004912 -0.021771 -0.491535 0.503651 -0.315409 0.031236 0.328521 -0.264316 -0.191875 0.681480 -0.782049 0.641517 0.380847 0.347847 0.043671 0.636368 -0.315160 -0.576198 -0.348432 0.281816 -0.623102 -0.564527 0.902504 0.016528 -0.069687 -0.144267 0.006254
月 0.282719 0.152989 0.277989 -0.409890 -0.100124 -0.562012 0.054540 -0.302864 0.340999 0.332058 -0.036765 -0.328907 -0.370917 0.120894 -0.257750 -0.695093 0.201110 -0.020349 -0.200095 0.133917 0.161953 1.409806 0.930571 0.444746 -0.639931 0.444198 -0.495933 0.229397 -0.739602 -1.055825 -0.318828 1.522916 0.004156 0.343912 0.006537 0.388347 0.600495 0.239206 -0.028486 0.552802 -0.278926 0.373066 -0.426863 0.489640 -0.323016 -0.080631 -1.074515 -0.507549 0.665036 0.391942 1.328556 1.073624 -0.177601 -0.018996 0.348987 0.525943 0.005171 -0.306071 -0.512433 -0.289054 0.337168 -0.289363 -0.288751 -0.316822 -0.342586 0.074602 -0.808880 0.993347 -0.027332 0.564592 -0.264008 0.547095 0.713716 0.306165 -0.489996 0.403748 -0.368705 -0.457879 0.797570 -0.495398 0.355519 1.163623 0.563982 -0.054508 -0.902991 -0.064551 -0.314185 -0.710030 -0.444326 -0.638333 0.392305 -0.397702 -0.002391 -0.020851 0.107497 0.182052 0.570882 -0.650250 0.326903 -0.717934 0.280774 -0.278874 0.623692 -0.195263 -0.542201 -0.003978 0.787480 0.940788 0.495447 0.390056 0.547322 -0.223878 -0.775611 0.423554 0.227717 0.880800 0.295964 0.302781 0.979723 -0.007546 -0.030809 -0.124463 1.068195 -0.250004 -0.743149 0.287037 -0.627813 0.902924 -0.019548 -0.008901 0.716885 -0.080450 -0.078079 -0.380885 0.595999 -0.006114 0.456368 0.310282 -0.593804 -0.097739 -0.533996 0.129195 -0.206451 0.854950 -0.736590 -0.373819 -0.171245 0.822648 0.022098 -0.041053 0.133135 -0.221786 -0.370930 -0.073366 -0.076312 -0.141111 0.754814 0.311759 -0.487563 0.148415 -0.327943 1.227186 1.104577 0.584394 0.866440 -0.155514 0.571861 0.225291 -0.596585 0.341632 -0.075625 0.459881 0.492822 -0.856172 -0.467885 1.009984 -0.509890 -0.289279 0.198226 0.779081 -0.505615 0.262267 -0.692102 -0.631975 0.688475 -0.343992 0.473175 0.662025 0.846393 -0.026749 -0.032720 -0.642198 -0.022452 -0.492827 0.074650 -0.200605 0.142318 -0.030474 -0.021498 -0.059870 -0.558756 -0.568076 -0.125238 0.087478 -0.520128 -0.406735 -0.211258 -0.481394 -0.664873 -0.277975 0.412186 0.010863 -0.240466 -0.070284 0.405342 0.066035 -1.461064 -1.013241 -0.072386 0.634662 0.716899 -0.185302 -0.030074 -0.113286 0.790407 -0.334036 0.145996 0.457602 0.061558 0.303261 0.445127 0.128329 0.371765 0.705406 -0.924010 0.312624 -0.264194 0.308366 0.098177 0.532446 -0.651350 0.452719 -0.764295 -0.191709 0.274529 -0.663116 -0.442445 -0.332471 -0.566812 0.441788 0.991128 0.062422 0.332908 0.114607 -0.790051 0.140813 -0.078728 0.221595 -0.313071 0.172639 0.681872 0.530355 -0.548417 0.459590 -0.557620 -0.267441 0.221526 -0.000649 0.288101 0.463587 -0.214246 0.953989 -0.281264 0.129526 -0.206172 -0.171313 -0.152700 0.777293 -0.521110 0.491064 0.467925 0.892538 -0.254171 -0.500852 0.145604 -0.039736 -0.016100 0.545795 -0.154586 0.515680 0.105515 -0.214572 -0.341872 0.009775 0.009587 1.174931 -0.101082 0.310132 -0.541959 -0.693216
元 0.306025 0.322643 1.077880 -0.600908 0.172935 -0.737794 0.255050 -0.157029 0.342953 -0.706849 -0.381400 0.061180 -0.953065 -0.095882 -0.368000 0.420154 0.951357 0.767554 -0.143634 -0.126899 0.307859 -0.300399 1.556839 0.364413 0.250135 -0.566936 -0.995738 -0.113883 -1.406691 -1.014822 -0.687600 1.741928 -0.615409 -0.387166 0.027234 1.281103 0.351747 -0.227581 -0.232469 0.271432 -0.415515 0.631718 -0.894145 0.733428 -1.230473 0.682133 0.075777 0.649245 0.289526 -0.467158 0.919363 0.351625 0.222911 0.504691 -0.181684 -0.378320 -0.687206 0.148117 0.729295 -0.105381 0.695989 -0.754219 -0.440530 0.118117 -0.883886 0.492664 0.079025 1.197542 -0.610626 0.157875 -0.443919 0.746728 1.468872 -0.575745 -0.470869 0.846473 -0.617876 -0.127038 0.003786 -0.001282 0.176192 0.778630 -0.308934 -0.249084 -0.301070 -0.087675 0.103930 -0.340054 0.051219 0.025974 0.446019 -0.389392 0.933938 0.164278 0.435473 0.114423 0.302819 -0.724311 -0.338384 -0.036844 0.040301 -0.075052 0.496395 0.007586 -0.240168 -0.644824 -0.383305 -0.000070 -0.135047 -0.441491 -0.397919 -0.251733 0.320516 -0.316922 -0.176393 0.178300 0.372434 0.236957 -0.165925 0.870158 -0.275455 -0.283281 0.730676 0.013333 0.467073 -0.715855 1.007568 0.391907 -0.500897 1.030966 0.698415 0.478425 -0.691962 0.789402 0.111583 0.636666 -0.887124 -0.272008 -0.584863 -0.175891 0.452481 0.355332 0.007005 0.805152 -0.325006 -0.249743 0.031626 0.073259 -0.453138 0.277470 -0.478755 0.575895 0.355874 0.035219 0.079009 0.054281 0.201341 -0.105456 0.624423 0.419490 0.422743 0.886797 0.058878 0.218280 0.783062 -0.278714 0.244279 0.244416 -0.558108 -0.538741 -0.626169 0.796777 0.313345 -0.594111 -0.612010 0.543808 0.326743 -0.068215 -0.284259 0.111663 -1.129291 0.464670 0.355996 0.317091 0.101179 -0.515349 -0.268207 1.036510 0.211519 0.545480 -0.140292 -0.338422 0.350183 0.160976 -0.810460 -0.581156 -0.083485 0.175628 -0.395788 -0.154374 1.656540 -0.149448 -0.250864 0.027195 -0.000960 0.097104 -0.385608 0.388430 -0.219321 0.263330 -0.339932 0.169096 -1.619655 -0.260263 0.345896 0.407777 -1.409953 -0.595680 -0.820713 -0.197955 0.132537 0.609146 0.432836 -0.071146 0.582325 -0.670098 0.034255 -0.291316 0.500515 1.061890 0.178922 -0.471838 -0.328243 -0.209035 -0.135114 -0.890339 -0.736016 -0.594963 -0.932622 -0.414669 0.308166 -0.405150 0.466669 0.111795 0.032414 -0.055596 -0.688865 -0.334978 -0.041121 0.252767 0.603890 0.051988 0.886804 -0.175748 0.236281 0.094143 0.452659 0.917740 0.303790 -0.453451 0.630142 -0.170290 -0.083950 0.310024 -0.560198 -0.581944 0.530984 -0.194871 -0.052266 0.184277 -0.055074 -0.165853 -0.889973 -0.130048 0.556185 0.093815 0.077349 0.786506 0.017906 -0.209109 -0.610210 0.882594 0.242138 0.931476 -0.261045 -0.497909 0.045525 -0.391315 -0.737701 0.693279 0.625252 0.169180 -1.320762 0.212941 0.034001 1.308421 0.299258 0.197882 -0.742295 -0.070419
1 0.888034 -0.269472 0.416726 -0.396031 0.035506 -0.622379 -0.440201 -0.753203 0.398985 0.105516 0.040577 0.550698 -1.196764 0.480859 -0.343003 -0.207138 0.024000 0.383471 -0.265794 0.036858 -0.098424 -0.568933 0.755142 -0.054576 -0.848028 -0.005950 -0.107111 -1.310041 -0.301031 -0.132762 -0.114449 1.529382 0.480697 -0.317541 0.098669 0.736248 0.507810 0.586016 0.222051 -0.894592 -0.685444 0.937100 -0.727683 0.896437 0.331013 -0.252685 -0.648829 0.198743 0.434566 -1.018625 0.604905 0.406292 -0.378488 -0.145570 0.143134 0.145109 0.082009 -0.334453 -0.308911 -0.548263 -0.306763 -0.239198 -0.025372 -1.286177 -0.418609 -0.941808 0.073460 0.799592 -0.180774 0.470524 -0.339510 0.074398 1.134824 -0.534680 -0.852488 0.674400 -0.772466 0.522004 0.311741 0.318874 0.701483 0.355505 0.916747 -0.391279 -0.043185 -0.035573 0.124245 -0.511008 -0.028164 0.396926 0.310719 -0.058037 0.415305 -0.124527 0.795043 -0.048217 -0.204688 -0.545567 -0.332399 0.533343 -0.196562 -0.587495 0.613407 0.339769 -1.312807 -0.425666 0.550779 -0.078751 0.106345 -0.597031 0.303927 0.367110 -0.479566 -0.280796 -0.077603 0.089361 0.057174 0.680617 -0.232158 -0.380329 -0.353017 0.261309 0.214358 -0.436605 0.726974 0.686759 -0.190239 0.569367 -0.357779 0.217572 0.577694 -0.200133 0.319214 -0.238840 0.142662 0.019461 -0.504774 -0.493205 0.434241 -0.077088 0.492905 0.143735 -0.385636 0.110834 -0.760336 -1.083028 0.465417 0.823979 -0.218171 0.331168 -0.171310 -0.281734 0.110895 0.805839 0.391114 -0.155342 0.366512 -0.215544 0.573037 0.302591 -0.715427 1.287327 -0.233937 1.247682 0.546919 -0.104151 0.185894 0.783857 -0.716013 0.745546 -0.484780 -0.330844 0.012054 -0.701770 -0.930120 0.304442 -0.101639 0.141293 0.213534 1.119318 -0.159903 0.034759 -0.237266 -0.027194 1.521587 0.373965 -0.609322 1.010848 0.325201 -0.254094 -1.082284 0.966625 0.455843 -0.249176 -1.112331 -0.351975 0.052057 -0.396617 0.975547 -0.350484 -0.519251 -0.116380 0.263111 -0.679581 -0.509177 -0.611612 -1.014958 -1.130154 -0.559550 -0.558102 0.844820 -0.233135 -1.107969 -0.443999 0.112733 -0.204075 -0.698707 -1.276953 0.029345 0.273015 0.776428 0.283971 1.089718 -0.201296 -0.199805 -0.994360 0.001024 0.000260 0.042881 0.442563 -0.074997 -0.225658 -0.127922 -0.043150 -0.446940 0.630705 0.122018 0.889063 -1.137543 0.602873 -0.074088 0.591213 -0.055548 -0.744662 -0.415526 0.587157 -0.649435 -0.512667 -0.191861 0.295901 1.263322 -0.303088 0.406485 0.562705 -0.070011 0.483705 -0.296791 -0.003154 0.232777 -0.667530 0.346206 -0.297299 -0.304221 -0.251610 -0.225808 -1.008746 1.032771 -0.805736 0.350475 0.072382 1.180248 0.153301 -0.149750 -0.278651 -0.837156 -0.453207 0.156093 -0.189694 -0.123239 -0.640496 0.453016 -0.009482 -0.390708 -0.866602 0.362385 -0.031449 0.210468 0.231915 -0.285665 0.196669 0.176305 0.104762 -0.411070 -0.084570 -0.912936 0.740259 1.085935 0.331030 -0.355904 -0.241756
会议 1.771033 0.176118 0.379083 -1.038539 -0.159619 -0.810625 -0.211994 0.767950 0.113235 0.995838 0.142457 0.287684 -0.035386 0.464643 -0.773942 -0.403471 0.569797 0.253740 0.426339 0.071171 -0.311686 0.865899 0.530907 -0.161541 -0.488560 -0.129126 -0.448615 -0.926315 -0.088304 -0.902380 -0.259576 0.434504 -0.757882 -1.191447 0.162485 0.108723 0.197529 0.581506 -1.198933 0.218464 0.503901 1.417884 -0.579565 0.713137 -0.258934 -0.476144 -0.548408 -0.012146 0.578107 0.107231 1.058120 0.882515 0.756237 -0.384693 0.165540 0.411003 0.348020 -0.023861 -0.259647 0.186697 0.023265 0.225137 -0.692051 -0.037526 0.435802 -0.293929 -1.062105 -0.157920 0.217099 -0.734419 -0.667745 -0.096092 0.113397 -0.059034 0.295290 0.334193 -0.308422 -0.613945 0.205872 -0.658714 0.406285 0.736497 -0.066900 -0.202817 -0.440141 0.164648 -0.142157 -0.441622 -0.122423 0.034315 0.195609 0.276887 1.015851 0.545567 -0.872668 -0.008203 -0.217614 -0.203448 -0.353687 0.261102 0.208269 -0.262863 0.746820 -0.594605 -1.031103 0.288283 0.015618 0.884347 -0.138220 0.333664 0.230371 0.605099 0.317958 0.915120 -0.128864 0.209937 0.466525 0.142983 0.987952 -0.338007 -0.503637 -1.041105 0.585695 0.133922 -0.283604 -0.389145 -0.513809 0.927723 0.178702 -0.179315 0.077576 -0.665182 -0.031870 0.141394 0.240645 0.393327 -0.711886 0.200745 -0.061365 -1.052649 0.527113 -0.460608 0.246856 -0.320261 -0.142330 -0.609020 0.796626 0.201946 -0.246875 0.225949 0.785565 -0.603375 0.355625 -0.179703 -0.322463 0.631364 -0.112776 0.125427 -1.077946 -0.163022 0.206300 -0.000588 -0.276898 -0.698788 0.289042 -0.117832 -0.423394 0.628158 -0.965626 0.353545 0.179244 -0.491377 0.257370 0.089590 -0.951331 -0.250147 -0.444209 0.185769 -0.132900 -0.006498 0.112555 -0.392332 -0.012566 -0.171049 0.412430 -0.784530 0.597930 0.654582 0.305766 -0.196717 -0.108986 -0.882588 0.133782 0.114576 -0.491082 0.697926 0.155321 0.534372 0.424671 0.057960 -0.566005 -0.098149 0.216918 -0.223675 0.078227 0.377749 -1.097760 -0.621768 -0.174247 -0.364606 1.090865 -0.314739 -1.034864 -0.132283 -0.111963 0.465461 -0.824262 -0.782546 0.374973 -0.286167 0.011139 -0.383112 0.354855 -0.439314 0.381809 -0.474764 0.159545 -1.221441 0.666809 -0.709793 0.239210 -0.138918 0.590115 0.215989 0.275636 0.433601 -0.156793 -0.382559 -1.424089 -0.417850 -0.116381 -0.344665 0.489679 0.626358 0.052841 0.577256 0.230637 -0.600817 -0.635985 1.089865 0.722713 0.494734 -0.428877 0.334609 -0.017065 0.336209 0.419211 1.363705 0.044493 0.188291 0.569913 -0.441831 0.071900 0.495701 -1.016109 0.061804 -0.363644 -0.767242 0.558061 0.567890 -0.662831 0.129088 -0.273309 -0.404180 -0.102150 -0.352034 0.208647 -0.281785 -0.134028 0.287623 -0.592395 -0.741625 -0.765389 -0.062867 1.148501 -0.574797 0.111045 -0.296249 0.241325 -0.732370 0.038897 -0.928541 0.283815 -0.379969 -1.189722 0.377329 0.206329 -0.463671 -0.644053 -0.356224
日 0.629690 0.297326 0.232061 0.184713 0.456603 -1.360426 -0.161957 -0.327643 0.379252 0.329272 -0.132093 -0.592853 0.369421 1.037765 0.194027 -0.582120 -0.030273 -0.433952 -0.491450 0.534424 0.298695 0.749407 0.961532 -0.013265 -0.160156 0.327203 0.178737 0.260298 0.023090 -0.661755 -0.069589 1.252100 0.274427 0.240321 0.676330 0.472751 0.397922 0.450867 -0.418843 0.434151 -0.318744 0.347155 -0.379078 -0.420713 -1.057411 0.168308 -0.487032 -0.069009 0.791306 0.176923 0.595739 0.527158 -0.328227 -0.121685 0.394574 0.216304 -0.410889 -0.465544 -0.511710 -0.106813 0.241581 -0.421071 -0.769960 0.508721 -0.518993 -0.582130 -0.122278 1.356947 -0.472147 0.219176 -0.637373 -0.379232 0.783050 -0.213542 -0.539033 0.326720 -0.806085 -0.636173 0.789530 -0.457049 0.013086 0.821359 0.303692 0.398572 -0.555042 -0.129366 -0.750004 -0.814094 -0.069134 -0.530091 0.515363 -0.586577 0.416111 -0.013937 -0.207776 0.584663 -0.013731 0.084024 -0.081933 -0.352394 0.082810 -0.485991 1.068765 -0.357893 -0.397341 -0.211468 0.420876 0.739444 0.097145 -0.032144 0.608081 0.445486 -0.453803 0.010084 0.063489 0.685313 0.311570 0.270301 0.373394 -0.579335 0.392215 -0.846188 1.196801 -0.524513 -1.024028 0.281200 -0.396088 0.811682 0.015575 0.444197 1.280806 -0.460657 -0.160046 -0.448320 0.348771 0.186178 -0.179169 -0.506007 -0.467586 0.232892 -0.132603 0.303326 0.532450 0.307448 -0.503961 -0.096622 -0.799405 0.108548 0.766091 -0.692119 -0.166302 -0.499754 -0.297644 -0.890542 -0.026685 -0.066082 0.513476 -0.234157 0.065489 0.236738 -0.698255 1.164697 0.689472 1.031871 0.946376 -0.528323 0.608606 -0.244125 -0.698136 0.663297 -0.315753 -0.024955 0.280253 -0.422331 -0.580495 0.018098 -0.057022 -0.306539 -0.084533 0.808313 -0.383531 0.320171 -0.399376 -0.477695 0.049535 -0.240776 0.169495 1.111698 0.712204 -0.209212 -0.668397 -0.478881 1.045714 -1.103660 0.234638 -0.466507 -0.287628 -0.133884 -0.486556 -0.456381 -0.277966 -0.006142 -0.013448 -0.202422 0.082728 -0.331743 -0.455172 -0.096092 -0.726063 -0.445307 0.451012 -0.380199 -0.300324 -0.068178 0.028520 0.143256 -0.972176 -1.223306 -0.187028 0.516180 -0.013776 -0.332469 -0.086373 -0.002370 0.777023 -0.041362 0.118836 -0.213045 0.513897 0.059171 0.371966 0.248342 -0.023563 0.700690 -0.921861 -0.051847 -0.109556 0.310796 -0.321488 0.309024 -0.368767 0.219925 -0.691066 -0.023356 0.030746 -0.452083 0.313607 -0.559715 -0.557504 0.528424 0.607529 -0.136493 0.152444 0.168481 -0.786771 0.006343 0.041056 0.642219 -0.609352 0.189559 0.682013 -0.063358 -0.391703 0.779091 -0.527853 0.707578 0.406660 0.103047 0.234198 0.038635 0.302829 1.064635 0.117944 -0.125693 -0.097658 -0.065264 0.228236 0.545942 0.338116 0.127504 0.171728 0.730071 -0.496038 0.158599 0.477939 0.339338 0.073317 0.780151 -0.409706 -0.080721 0.052942 0.037625 -0.642805 -0.008267 0.050677 1.016556 0.096010 0.497020 0.160714 -0.086144
12 0.060987 0.220731 -0.132685 -0.820905 -0.146994 -0.353599 -0.018432 -0.725461 0.737936 -0.019994 -0.154258 -0.419405 -0.338631 -0.291629 -0.022446 -0.251117 0.267038 0.434467 0.078770 -0.025413 0.500651 0.854019 0.395952 -0.051031 -0.435698 0.237282 -0.583161 0.168157 -0.601435 -1.348934 -0.164784 1.616810 -0.478824 -0.017075 -0.366454 0.365084 0.612911 0.161186 0.276902 0.375643 -0.112530 0.241358 -0.396695 -0.194163 -0.237141 0.697846 -0.526700 -0.531520 0.772570 -0.322104 1.315113 0.933782 -0.300440 0.198085 0.157843 0.812475 0.187118 -0.254738 0.139663 -0.073107 0.528683 -0.647160 -0.423146 0.006624 -0.894147 -0.726237 -0.881079 0.825501 0.080627 -0.561961 -0.475129 0.047358 0.167624 0.359150 -0.428893 -0.098055 -0.787332 -0.403023 0.472614 -0.063268 0.086656 0.310264 0.610745 -0.221438 -0.421589 0.778696 0.256653 -0.560248 -0.649633 -0.102870 0.586092 0.003949 0.076587 -0.567913 0.704371 0.364411 0.761299 -0.051966 0.385057 -0.367773 0.355770 -0.647424 0.923446 -0.051832 -0.543764 0.588581 0.554076 0.719280 0.339640 0.226296 0.472695 0.397156 -0.489635 0.773813 -0.117013 0.493147 0.093860 0.815739 0.779326 0.310038 -0.783332 -0.620891 0.599204 0.011149 -0.107909 0.348672 -0.498058 0.057396 -0.031033 0.168456 0.991581 -0.478983 -0.781825 -0.271430 0.955011 -0.478941 0.302897 -0.093343 -1.225087 0.255701 -0.017678 -0.452325 0.377625 0.635914 -1.060914 -0.345057 -0.322616 0.805557 0.755067 0.543725 0.030418 -0.496267 -0.373508 -0.095644 -0.015342 0.230422 0.639305 0.210231 0.053320 0.272379 0.220056 0.767136 0.343294 0.690162 1.038597 -0.527260 0.498455 0.133132 -0.857316 -0.088142 -0.121594 0.524317 0.324666 -0.600952 -0.612691 1.126837 -0.052147 -0.754994 -0.371588 0.592034 -0.199333 -0.271918 -0.781767 -0.777349 0.545457 -0.369653 0.403855 0.507788 0.341316 0.256601 -0.652804 -0.737112 -0.248066 -0.641719 0.033328 -0.249865 0.328787 0.311992 -0.004281 0.216633 -0.081194 -0.126300 -0.272642 -0.001156 -0.393779 -0.830402 0.065871 -0.295060 -0.708441 -0.639450 -0.084257 0.074900 -0.591158 -0.031511 0.261493 -0.149527 -1.486934 -0.835603 -0.202950 0.765206 0.159304 -0.304259 0.363361 -0.466972 1.143983 -0.557980 -0.238127 -0.099868 0.267013 0.080638 0.565594 -0.164587 0.111982 0.391523 -0.656658 0.514244 -0.385658 -0.082066 -0.306449 0.810565 -0.694959 -0.296500 0.137355 -0.246647 0.456403 -0.376321 0.404851 -0.319968 -0.270184 0.835729 0.550599 0.198025 1.204497 -0.237713 -0.441482 -0.020244 0.109327 -0.164325 -0.587616 0.205860 0.452751 0.599642 -0.033190 0.569953 0.012436 -0.112856 0.127171 -0.183683 0.785418 0.581380 -0.613472 0.544212 0.165229 0.089152 -0.788505 -0.047272 -0.276138 -0.010708 -0.105783 0.323865 0.158936 0.329226 -0.172630 0.240504 0.143898 0.473886 -0.170357 0.875988 -0.070156 0.038928 0.481665 0.419615 0.238390 0.543345 0.468181 0.671512 -0.106844 -0.193693 -0.246399 -0.186887
2 0.361149 0.704917 0.291457 -0.080728 -0.572933 -0.589067 -0.793803 -1.000979 0.820175 0.295928 0.184207 0.885769 -0.399009 0.637581 0.318221 -0.022845 0.259705 -0.114552 -0.054940 0.182685 0.221006 -0.218122 0.865297 -0.441181 0.185419 0.447350 -0.213860 -1.769530 -0.623723 -1.066396 0.099660 1.458072 0.450007 -0.789303 -0.121537 0.312682 0.246968 0.901696 0.033775 0.563942 0.521539 0.790706 -0.875751 0.932463 -0.507096 -0.130464 1.109031 -0.469966 0.884996 -0.271595 0.670886 0.471352 -0.666085 0.080534 0.530477 0.371931 -0.060190 -0.320351 0.219629 -0.268545 0.640168 -0.545242 0.073721 -0.646921 0.468086 -0.719434 -0.026094 0.861088 0.010164 0.028802 -0.204849 0.493036 0.357690 -0.472914 -0.582728 0.376733 0.621382 -0.667704 -0.412258 -0.081961 0.306849 -0.197246 0.451879 -1.593393 0.159934 -0.159739 0.726716 0.039497 -0.330342 -0.193373 -0.045038 -0.785062 0.898486 -0.217732 1.028231 0.260610 0.865236 -0.474029 -0.726025 -0.554919 -1.216791 -0.570384 0.724477 -0.346603 -0.662520 0.868426 0.021664 -0.019214 0.271993 -0.176794 0.194711 -0.084973 -1.092003 -0.006232 -0.776445 -0.021545 0.000348 -0.076073 0.193496 0.259513 -0.696651 -0.119337 0.141526 0.255560 0.426479 0.253425 -0.899710 0.643363 -0.486986 0.184396 1.035831 0.141304 0.019429 0.155365 0.002902 -0.109107 0.055377 -0.024775 -0.693244 0.206938 0.179723 -0.798811 0.141440 -0.525572 -0.577044 -0.878470 0.477901 0.946455 -0.141953 -0.503441 -0.478369 0.069927 -0.136817 0.606908 -0.076950 0.324349 0.192879 0.114894 -0.919087 0.352348 -1.414676 0.916139 0.328463 1.186773 0.745050 -0.434486 -0.267733 -0.199708 -0.534591 0.888049 -0.429322 0.363039 -0.040896 -0.322206 -0.337510 -0.026441 -0.142639 0.213708 -0.124947 0.622512 -1.049125 -0.706038 -0.208899 0.447170 0.536233 -0.231709 0.489767 0.514644 0.048299 -0.557416 0.273464 -1.164850 0.695038 0.409650 -0.814939 -0.453800 0.442305 -0.609455 0.602072 0.006255 -0.069054 -0.450125 0.125254 -0.478271 -0.413480 -1.055180 -0.893434 -0.556728 -1.093960 -0.768353 0.271325 -0.388704 -0.630777 0.779149 -0.211849 -0.288888 -0.145470 -0.570022 -0.030645 -0.831691 0.284689 0.429069 0.362861 -0.749234 0.465644 0.248452 -0.246630 -0.053088 0.302568 0.058695 -0.025171 0.224125 -0.313906 0.595948 -0.747329 0.141962 -0.211617 0.277275 -0.510860 0.348968 0.048324 -0.095582 0.024356 -0.532523 -0.415193 0.493446 0.292916 -0.496475 0.292169 -0.555822 0.702531 -0.081132 -0.078505 0.356687 0.408023 0.532309 -0.100873 0.750387 0.157968 -0.626308 0.372636 0.901865 0.156366 -0.398677 -0.315714 0.024024 -0.440340 0.041518 0.937736 0.692113 -0.201351 -0.068569 -0.755074 -0.320450 -0.319768 -0.102927 0.009273 0.043572 -0.163320 -0.071635 0.635202 0.503277 0.106900 -0.362000 0.350847 -0.668718 -0.516442 -0.376715 -0.380673 0.517315 0.964409 -0.042094 -0.786081 -0.959971 -0.023261 1.323840 1.026073 -0.052168 -0.350861 0.062560
年 -0.025680 -0.148341 0.260403 0.435279 0.340753 -0.370165 0.692524 0.405896 0.703198 0.419036 0.065059 1.252715 -0.546617 0.288202 0.012702 0.090569 -0.643207 -0.240639 0.319901 -0.293442 0.487747 1.345064 0.846493 0.117547 -1.066390 0.662450 0.008567 -0.178598 -0.069384 -0.303476 -0.419415 0.074236 -0.287813 -0.494250 -1.147110 -0.279624 0.231568 0.129380 -0.133671 0.771361 0.169357 0.685386 -0.558743 -0.012061 -0.356053 0.066274 -0.478152 0.360300 0.584327 -0.111067 0.741198 0.946980 -0.227466 0.026979 0.199299 -0.356406 0.466501 -0.262172 -0.271103 -0.312378 0.475935 0.135873 -0.576739 -0.376040 0.230084 0.311886 -0.885410 -0.017326 0.172162 -0.166069 -0.669513 -0.079418 0.324018 0.229297 0.349141 1.200803 -0.836550 -0.131899 0.847725 0.495316 0.118726 0.463500 0.874937 -0.116077 -0.785283 0.009747 -1.198582 -0.528437 -1.039171 0.715792 0.459527 -0.346323 0.090036 0.272467 -0.146408 -0.137629 0.630577 0.404533 0.325585 -0.837493 0.497690 -0.118868 0.341744 0.342129 -0.100156 0.656184 0.002508 0.282706 0.801364 -0.268771 0.100605 0.126436 -0.148326 0.836686 0.522442 0.337463 0.759250 0.411933 0.206825 -0.341759 -1.007716 -1.160924 0.417559 0.168041 -0.196686 -0.114311 -0.240378 0.511523 -0.102376 0.243313 0.841565 -0.114063 -0.525006 -0.811578 -0.088705 0.243166 0.499764 -0.087099 -0.166266 -0.083989 -0.240918 0.481979 -0.541293 0.408957 -0.772534 -0.699110 0.168422 0.308715 -0.136194 0.043910 -0.489641 -1.030214 0.028125 0.707278 -0.245171 0.421922 -0.013471 0.309093 -0.260784 -0.311308 0.071559 0.550136 0.403435 0.467421 1.336632 0.114131 -0.127869 0.211865 -0.463383 0.807332 -0.462692 -0.031325 0.557738 -0.334642 -0.955832 1.272109 -0.304985 -0.560109 0.276666 0.388646 -1.179766 0.460541 -1.091042 -0.190026 0.188740 -0.575042 -0.085519 0.385787 0.031602 -0.232948 -0.306679 -0.166243 0.281174 -0.706293 -0.701480 -0.668203 -0.032880 0.267294 -0.275263 0.358886 -0.889863 -0.086739 0.058511 -0.305190 0.250048 -0.219450 -0.859118 -0.484297 -0.703766 -0.833164 0.497965 -0.961938 -0.814127 0.056959 -0.110192 0.812637 -0.936772 -0.882466 -0.594758 0.620133 -0.045738 -0.646158 0.377140 0.196992 0.951233 0.108737 0.363643 0.312669 0.062553 0.577432 0.085247 0.433714 -0.561889 1.315280 -0.046503 1.064645 -0.535019 -0.173368 -0.653338 0.681322 -0.250618 0.710895 0.240201 -0.413712 -0.050010 0.009633 -0.610238 -0.342928 0.126470 0.558831 0.700774 0.322080 0.095641 0.450095 -0.186828 0.389571 0.042733 -0.028652 0.072356 -0.591279 0.306309 1.063946 -0.136931 -0.370101 -0.133079 -0.435311 -0.852584 0.450015 0.808773 0.596418 -0.594216 0.450295 0.264785 0.314351 -0.079821 0.326998 -0.191889 0.554631 0.185501 0.550335 0.010618 0.740444 -0.118180 -0.088854 0.899637 -0.636072 -0.017279 -0.687941 -0.471457 0.413444 0.003071 -0.017811 0.194908 -0.023812 -0.580133 0.906154 -0.002010 0.835716 -0.461930 -1.249504
请

3、Gensim加载GloVe训练的词向量

from gensim.models import KeyedVectors
from gensim.scripts.glove2word2vec import glove2word2vec
# 输入文件
glove_file = 'data/glove/vectors.txt'
# 输出文件
w2v_file = 'data/glove/w2v.txt'
# 开始转换
glove2word2vec(glove_file, w2v_file)
# 加载转化后的文件
model = KeyedVectors.load_word2vec_format(w2v_file)   #该加载的文件格式需要转换为utf-8
print(model['时间'])  #这个就和Word2Vec训练的模型使用方法一样了

三、Doc2Vec

1、Doc2Vec介绍

Word2Vec表示的词向量不仅考虑了词之间的语义信息，还压缩了维度。但是，有时候当我们需要得到Sentence/Document的向量表示，虽然可以直接将Sentence/Document中所有词的向量取均值作为Sentence/Document的向量表示，但是这样会忽略单词之间的排列顺序对句子或文本信息的影响。

Doc2vec是在Word2vec的基础上做出的改进，它不仅考虑了词和词之间的语义，也考虑了词序。

Doc2Vec论文

2、Doc2Vec模型

Doc2Vec有两种模型，分别为：句向量的分布记忆模型（PV-DM: Distributed Memory Model of Paragraph Vectors）和句向量的分布词袋（PV-DBOW: Distributed Bag of Words version of Paragraph Vector）。

DM模型在给定上下文和文档向量的情况下预测单词的概率。即在训练时，首先将每个文档ID和语料库中的所有词初始化一个K维的向量，然后将文档向量和上下文词的向量输入模型，隐层将这些向量累加（或取均值、或直接拼接起来）得到中间向量，作为输出层softmax的输入。在一个文档的训练过程中，文档ID保持不变，共享着同一个文档向量，相当于在预测单词的概率时，都利用了这个句子的语义。

DBOW模型在给定文档向量的情况下预测文档中一组随机抽样的单词的概率。

注：Doc2vec的DM模型跟Word2vec的CBOW很像，DBOW模型跟Word2vec的Skip-gram很像。Doc2Vec为不同长度的段落训练出同一长度的向量；不同段落的词向量不共享；训练集训练出来的词向量意思一致，可以共享。

3、Gensim实现Doc2Vec

from gensim.models.doc2vec import Doc2Vec, TaggedDocument
import pandas as pd
import numpy as np
# 构建Doc2vec模型，获得句子向量
def get_sentence_vec(datasets):
# gemsim里Doc2vec模型需要的输入为固定格式，输入样本为[句子，句子序号]
documents = [TaggedDocument(doc, [i]) for i, doc in enumerate(datasets)]
# 初始化和训练模型
model = Doc2Vec(documents, vector_size=500, dm=1, window=4, min_count=5,epochs=50)
#model = Doc2Vec(vector_size=300, dm=1, window=4, min_count=5, epochs=50)
#model.build_vocab(documents)
#model.train(documents,total_examples=model.corpus_count,epochs=model.epochs)
model.save('data/w2v/doc2vec_model.pkl')  # 将模型保存到磁盘
# 获得数据集的句向量
documents_vecs = np.concatenate([np.array(model.docvecs[sen.tags[0]].reshape(1, 300)) for sen in documents])
return documents_vecs
if __name__=='__main__':
#准备数据
train_data = pd.read_csv('data/clean_data_train.csv', sep=',', names=['contents', 'labels']).astype(str)
test_data = pd.read_csv('data/clean_data_test.csv', sep=',', names=['contents', 'labels']).astype(str)
cw = lambda x: str(x).split()
train_data['words'] = train_data['contents'].apply(cw)
test_data['words'] = train_data['contents'].apply(cw)
datasets = pd.concat([train_data, test_data])
#doc2vec句向量训练和生成
documents_vec=get_sentence_vec(list(datasets['words']))
#加载训练好的模型
doc2vec_model=Doc2Vec.load('data/w2v/doc2vec_model.pkl')
#推断新文档向量
doc2vec_model.infer_vector(['绝望','快递','说','收到','快递','中奖','开心'])

四、Fasttext

1、Fasttext介绍

Fasttext可以实现高效学习单词表示和句子分类；Fasttext是一个快速文本分类算法，与基于神经网络的分类算法相比有两大优点。

Fasttext在保持高精度的情况下加快了训练速度和测试速度
Fasttext不需要预训练好的词向量，Fasttext会自己训练词向量

Fasttext论文 Fasttext源码 Fasttext官网 Fasttext原理和实现 Fasttext源码解析 Fasttext源码解析

2、Fasttext学习单词向量表示（Linux）

使用fasttext.train_unsupervised函数训练词向量模型

代码实现如下：

import fasttext
# Skipgram model
model = fasttext.train_unsupervised('data.txt', model='skipgram')
# cbow model
model = fasttext.train_unsupervised('data.txt', model='cbow')
print(model.words)   # list of words in dictionary
print(model['king']) # get the vector of the word 'king'
model.save_model("model_filename.bin")  #可以通过调用该函数来保存训练的模型对象save_model
model = fasttext.load_model("model_filename.bin") #加载训练好的模型

其中，data.txt是包含utf-8编码文本的训练文件；返回的model对象表示学习的词向量模型，可以使用该模型检索信息。

2、Fasttext实现文本分类（Linux）

使用fasttext.train_supervised函数训练文本分类器

代码实现如下：

# coding=utf-8
import pandas as pd
import numpy as np
import fasttext
from sklearn import metrics
# CSV文件转换为fasttext格式的文件
def fasttext_file():
train_data = pd.read_csv('data/clean_train_data.csv', sep=',', names=['contents', 'labels']).astype(str)
test_data = pd.read_csv('data/clean_test_data.csv', sep=',', names=['contents', 'labels']).astype(str)
train_rows = len(train_data)
test_rows = len(test_data)
with open('data/fasttext_train.txt', 'a') as f:
for i in range(train_rows):
f.write(train_data['contents'][i] + '\t' + '__label__' + train_data['labels'][i] + '\n')
f.close()
with open('data/fasttext_test.txt', 'a') as f:
for i in range(test_rows):
f.write(train_data['contents'][i] + '\t' + '__label__' + train_data['labels'][i] + '\n')
f.close()
if __name__ == '__main__':
# fasttext_file()
# 第一个参数是前面得到的 fasttex_train.txt ，第二个参数是将要保存模型的路径，默认会加上.bin
# label_prefix 就是标签或类别的起始符号
classifier = fasttext.supervised("data/fasttext_train.txt", "data/fasttext.model", label_prefix="__label__",dim=5000, epoch=100, min_count=10)
# 加载模型
classifier = fasttext.load_model("data/fasttext.model.bin", label_prefix="__label__")
# 测试模型 其中 fasttext_test.txt 就是测试数据，格式和 fasttext_train.txt 一样
result = classifier.test("data/fasttext_test.txt")
print(result)
print("准确率:", result.precision)
print("召回率:", result.recall)

其中，fasttext_train.txt是一个文本文件，每行包含一个训练语句以及标签。默认情况下，我们假设标签是以字符串为前缀的单词__label__

import fasttext
model = fasttext.train_supervised('fasttext_train.txt')
#我们还可以预测特定文本的标签
model.predict("Which baking dish is best to bake a banana bread ?")
#默认情况下，predict仅返回一个标签：概率最高的标签。您还可以通过指定参数来预测多个标签k：
model.predict("Which baking dish is best to bake a banana bread ?", k=3)
#如果要预测多个句子，可以传递一个字符串数组：
model.predict(["Which baking dish is best to bake a banana bread ?",
"Why not put knives in the dishwasher?"], k=3)
#model_filename.ftz将比一个小得多的尺寸model_filename.bin
model.save_model("model_filename.ftz")