自然语言处理文本相似度文本语义相似度计算

转载

桃太郎 2024-06-07 16:48:05

文章标签 自然语言处理文本相似度 tf-idf python 自然语言处理词向量 文章分类 NLP 人工智能

主要步骤流程：

文本预处理：分词

训练词向量

tf-idf提取关键词并写入文档前20个词

word2vec读取提取的关键词并转为词向量

相似度计算。

前面对word2vec的理论进行了介绍，接下来进入实战环节。实践主要分为三部分，word2vec训练，相似度计算以及训练结果展示三部分。

词向量的训练：

词向量的训练主要分为两部分：

1、对中文语料进行数据预处理；

2、利用gensim模块训练词向量。

中文语料处理：

在用语料库进行训练词向量之前需要对中文句子进行分词，这里采用jieba中文分词工具对句子进行分词。具体代码如下：

import GrobalParament
import jieba
import pandas as pd

#去掉回车换行
def delete_r_n(line):
    return line.replace("\r","").replace("\n","").strip()

#读取停用词
def get_stop_words(stop_words_dir):
    stop_words = []

    with open(stop_words_dir, "r", encoding=GrobalParament.encoding) as f_reader:
        for line in f_reader:
            line = delete_r_n(line)
            stop_words.append(line)

    stop_words = set(stop_words)
    return stop_words


#结巴精准分词
def jieba_cut(content, stop_words):
    word_list = []

    if content != "" and content is not None:
        seg_list = jieba.cut(content)
        for word in seg_list:
            if word not in stop_words:
                word_list.append(word)

    return word_list

#结巴索索引擎分词
def jieba_cut_for_search(content, stop_words):
    word_list = []

    if content != "" and content is not None:
        seg_list = jieba.cut_for_search(content)
        for word in seg_list:
            if word not in stop_words:
                word_list.append(word)

    return word_list

#清除不在词汇表中的词语
def clear_word_from_vocab(word_list, vocab):
    new_word_list = []

    for word in word_list:
        if word in vocab:
            new_word_list.append(word)

    return new_word_list

#文本预处理
def preprocessing_text_pd(text_dir, after_process_text_dir, stop_words_dir):
    stop_words = get_stop_words(stop_words_dir)
    setences = []
    df = pd.read_csv(text_dir)
    for index, row in df.iterrows():
        print(index)
        title = delete_r_n(row['title'])
        word_list = jieba_cut(title, stop_words)
        df.loc[index,'title'] = " ".join(word_list)
        setences.append(word_list)
    df.to_csv(after_process_text_dir,encoding=GrobalParament.encoding, index=False)
    return setences

#文本预处理第二种方式
def preprocessing_text(text_dir, after_process_text_dir, stop_words_dir):
    stop_words = get_stop_words(stop_words_dir)
    setences = []
    f_writer = open(after_process_text_dir,"w", encoding=GrobalParament.encoding)
    count = 0
    with open(text_dir, "r", encoding=GrobalParament.encoding) as f_reader:
        for line in f_reader:
            line_list = line.split(",")
            if len(line_list) == 2:
                line_list[1] = delete_r_n(line_list[1])
                word_list = jieba_cut(line_list[1], stop_words)
                setences.append(word_list)
                f_writer.write(line_list[0] + "," + " ".join(word_list) + "\n")
                f_writer.flush()
                count = count + 1
                print(count)
            else:
                pass
                # print(line)

    f_writer.close()
    return setences


if __name__ == "__main__":
    stop_words = get_stop_words(GrobalParament.stop_word_dir)
    setences = preprocessing_text(GrobalParament.test_set_dir, GrobalParament.test_after_process_text_dir, GrobalParament.stop_word_dir)
    # setences = preprocessing_text_pd(GrobalParament.train_set_dir, GrobalParament.train_after_process_text_dir, GrobalParament.stop_word_dir)
    #print(setences[:10])

最后将处理后的语料存入txt文档中，处理后的语料如下图所示：

自然语言处理文本相似度文本语义相似度计算_词向量

向量化训练：

利用gensim模块训练词向量。之前有详细介绍训练方法和参数设置，这里不再累赘。

import GrobalParament
import utils
from gensim.models import word2vec

#训练word2vec
def train(sentences, model_out_put_path):
    print("开始训练")
    model = word2vec.Word2Vec(sentences = sentences, size = GrobalParament.train_size, window = GrobalParament.train_window, min_count = 20)
    model.save(model_out_put_path)
    print("训练完成")

if __name__ == "__main__":
    # sentences = utils.preprocessing_text(GrobalParament.train_set_dir, GrobalParament.train_after_process_text_dir, GrobalParament.stop_word_dir)
    # train(sentences, GrobalParament.model_output_path)
    model = word2vec.Word2Vec.load(GrobalParament.model_output_path)
    vocab = list(model.wv.vocab.keys())
    for e in model.most_similar(positive = ['漏水'], topn = 10):
        print(e[0],e[1])
    print(len(vocab))