python 实现词向量的表示 包括词袋模型/TF-IDF/Word2Vec_将herryporter电子书作为语料库,分别使用词袋模型,tf-idf模型和word2vec模型进

以下分别是TF（词袋模型）/TF-IDF/skip-gram(Word2vec)的具体实现代码。

1. 词袋模型

from sklearn.feature_extraction.text import CountVectorizer
#语料库
train_x= ['build fails due publication-tests.xml build target','due to sb']
test_x =['build one to ']
#将文本中的词语转换为词频矩阵  选择前256个词 相当于词向量的维度是256维的
cv_ = CountVectorizer(max_features=256)
#计算个词语出现的次数  此类方法一般先fit拟合，再transform转换
X = cv_.fit_transform(train_x)
#输出语料库
print('corpus',train_x)
#输出词典
print('feature_names',cv_.get_feature_names())
#输出词汇
print('vocabulary_',cv_.vocabulary_)
#输出模型参数
print('params',cv_.get_params(deep=True))
#输出词频
print(X)
#查看词频结果
print(X.toarray())
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

结果：

（0，0） 1 表示第1句预料中，第0个单词，出现的次数

---

封装为函数

from sklearn.feature_extraction.text import CountVectorizer

def count(train_x,test_x):
    cv_ = CountVectorizer(max_features=256)
    # 使用训练数据train_x进行词向量的拟合，得到训练集的词向量表示结果data_train_tf，然后输出矩阵形式
    data_train_tf = cv_.fit_transform(train_x).toarray()
    # 对测试集数据test_x进行词向量的转换
    data_test_tf = cv_.transform(test_x).toarray()
    return data_train_tf,data_test_tf
1
2
3
4
5
6
7
8
9

2. TF-IDF

from sklearn.feature_extraction.text import TfidfVectorizer
# 理论同上
def tfidf(train_x,test_x):
    tr_ = TfidfVectorizer(max_features=256)
    data_train_tfidf = tr_.fit_transform(train_x).toarray()
    data_test_tfidf = tr_.transform(test_x).toarray()
    return data_train_tfidf,data_test_tfidf
1
2
3
4
5
6
7

3. Word2Vec

from gensim.models import Word2Vec
from gensim.models.word2vec import LineSentence
import numpy as np

# 获取skip-gram的词向量 一句话中所有词向量求均值得到该句的向量
def get_embed(skip_,data_path,dim_size=300):
    data_res=[]
    for s_ in LineSentence(data_path):
        vec_res = np.zeros(dim_size)
        for t_ in s_:
            try:
                vec = skip_[t_]  #不存在该单词
                vec_res += vec
            except:
                vec = np.zeros(dim_size)
                vec_res += vec
        data_res.append(vec_res/len(s_))
    return np.array(data_res)

# 输出skip-gram表示的词向量 词向量大小维度为300 retain表示是否覆盖之前的训练结果
def word2vec(train_x,test_x,soft,dim_size=300,retrain=True):
    data_path_train = 'F:/shiyan/bert/dataset/generate/'+soft+'/train_x.txt'
    save_path_train = 'F:/shiyan/bert/dataset/generate/'+soft+'/skip_.model'
    if retrain:
    	# 将训练数据train_x以一行一句话的形式存入txt文件中
        train_x.to_csv(data_path_train, sep='\t',index=False, header=None)
        # 训练skip-gram
        skip_ = Word2Vec(LineSentence(data_path_train),size=dim_size,window=10,sg=1)
        # 保存模型
        skip_.save(save_path_train)
        # 单词 向量的存储
        skip_.wv.save_word2vec_format('F:/shiyan/bert/dataset/generate/'+soft+'/skip_keys.model', binary=False)
    else:
    	# 若之前训练过，这里直接加载训练好的模型
        skip_ = Word2Vec.load(save_path_train)
    data_train_skip = get_embed(skip_,data_path_train,dim_size=dim_size)


    # 加入测试集再次训练
    data_path_test = 'F:/shiyan/bert/dataset/generate/'+soft+'/test_x.txt'
    save_path_test = 'F:/shiyan/bert/dataset/generate/'+soft+'/skip_latter.model'
    test_x.to_csv(data_path_test, sep='\t',index=False, header=None)
    # 对模型进行微调
    skip_.train(LineSentence(data_path_test),total_examples=skip_.corpus_count,epochs=skip_.epochs)
    skip_.save(save_path_test)
    skip_.wv.save_word2vec_format('F:/shiyan/bert/dataset/generate/'+soft+'/skip_keys_latter.model', binary=False)
    data_test_skip = get_embed(skip_,data_path_test,dim_size=dim_size)
    return data_train_skip,data_test_skip

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49