计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (16): 123-128.DOI: 10.3778/j.issn.1002-8331.2104-0203

• 模式识别与人工智能 • 上一篇    下一篇

基于句向量和卷积神经网络的文本聚类研究

贾君霞,王会真,任凯,康文   

  1. 1.兰州交通大学 电子与信息工程学院,兰州 730070
    2.国电甘肃新能源有限公司,兰州 730070
  • 出版日期:2022-08-15 发布日期:2022-08-15

Research on Text Clustering Based on Sentence Vector and Convolutional Neural Network

JIA Junxia, WANG Huizhen, REN Kai, KANG Wen   

  1. 1.School of Electronics and Information Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China
    2.Guodian Gansu New Energy Co., Ltd., Lanzhou 730070, China
  • Online:2022-08-15 Published:2022-08-15

摘要: 针对文本聚类时文本特征维度高,忽略文档词排列顺序和语义等问题,提出了一种基于句向量(Doc2vec)和卷积神经网络(convolutional neural networks,CNN)的文本特征提取方法用于文本聚类。首先利用Doc2vec模型把训练数据集中的文本转换成句向量,充分考虑文档词排列顺序和语义;然后利用CNN提取文本的深层语义特征,解决特征维度高的问题,得到能够用于聚类的文本特征向量;最后使用[k]-means算法进行聚类。实验结果表明,在爬取的搜狗新闻数据上,该文本聚类模型的准确率达到了0.776,F值指标达到了0.780,相比其他文本聚类模型均有所提高。

关键词: 卷积神经网络(CNN), Doc2vec, 文本表示, 文本聚类

Abstract: Aiming at the problems of the high dimensionality of text features in text clustering, and ignoring the order and semantics of document words, this paper proposes a text feature extraction method based on Doc2vec and convolutional neural networks(CNN) for text clustering. Firstly, use the Doc2vec model to convert the text in the training dataset into sentence vectors, fully consider the order and semantics of the document words. Then, use CNN to extract the deep semantic features of the text, solve the problem of high feature dimensions, and obtain the data that can be used for clustering text feature vector. Finally, use the [k]-means algorithm for clustering. The experimental results show that on the crawled Sogou news data, the accuracy of the text clustering model proposed in this paper has reached 0.776, and the F-score index has reached 0.780, which is improved compared to other text clustering models.

Key words: convolutional neural networks(CNN), Doc2vec, text representation, text clustering