基于卷积神经网络的文献分析

发布时间：2018-03-13 12:07

本文选题：卷积神经网络　切入点：word2vec　出处：《长春工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：在互联网上的文本信息量飞速增长的同时,人们对互联网上的论文文献进行查阅和整理所需要耗费的时间也越来越多。在这种背景下,利用文本聚类技术来对海量的文献进行分类整理,具有非常重要的应用前景与研究意义。文本聚类是文本挖掘的一项重要技术,能够广泛应用于文本挖掘与信息检索等方面,在大规模文本集的组织与浏览、文本集层级归类的自动生成方面都具有重要的应用价值。文本聚类的首要问题是如何将文本数据以数学形式表示出来,同时传统的文本聚类算法忽略了文本中单词之间的语义相关性,以及传统聚类算法存在聚类结果不稳定等问题。本文主要是针对以上问题对文本聚类进行研究。本文使用搜狗中文语料库、复旦大学中文语料库和哈尔滨工业大学中文语料库中的中文文本数据作为实验的数据集,通过利用word2vec工具对文本数据做词向量转化、利用卷积神经网络对文本数据进行特征提取以及利用基于K-means改进的KSDM聚类算法对文本数据进行聚类,达到对论文文献聚类的目的。本文主要工作如下:1、本文首先探讨了文本聚类算法的研究意义及国内外研究现状;分析了当前传统的文本聚类算法存在的不足;2、研究了常用的几种文本聚类算法,卷积神经网络的基本原理,以及词向量的转变和word2vec工具的基本原理。3、设计了基于卷积神经网络的文本特征提取方法。搭建了卷积神经网络模型,选取了卷积神经网络的各项参数。通过实验验证本文设计的基于卷积神经网络的文本特征提取方法的有效性。4、设计了基于k-means改进的KSDM聚类算法。在传统的K-means算法的基础上,针对其自身的不足提出了一种新的孤立点检测算法和一种新的聚类中心选取算法。实验结果证明了KSDM算法的有效性。5、在理论研究的基础上,将word2vec工具、卷积神经网络和KSDM聚类算法相结合,提出了基于卷积神经网络的文献分析框架。首先需要对文本数据进行分词、去除停用词、以及词向量转化等预处理过程,然后将得到词向量存储到向量矩阵中并输入到预先训练的卷积神经网络中来提取文本特征,最后,将得到的特征输入到KSDM聚类算法中进行聚类,已实现对测试文献聚类。实验结果验证,本文算法有效的提高了现有的文本聚类算法的准确度,并且具有较高的扩展性和灵活性。
[Abstract]:With the rapid growth of text information on the Internet, it takes more and more time for people to consult and organize the papers and documents on the Internet. Text clustering is an important technology in text mining, which can be widely used in text mining and information retrieval. The organization and browsing of large-scale text sets and the automatic generation of text set classification have important application value. The most important problem of text clustering is how to express the text data in mathematical form. At the same time, the traditional text clustering algorithm ignores the semantic correlation between the words in the text. And the traditional clustering algorithm has some problems, such as the instability of clustering results. This paper mainly focuses on the research of text clustering in view of the above problems. The Chinese text data in the Chinese Corpus of Fudan University and the Chinese Corpus of Harbin Polytechnic University are used as the experimental data sets, and the word vector transformation of the text data is done by using word2vec tool. Using convolutional neural network to extract the feature of text data, and using the improved KSDM clustering algorithm based on K-means to cluster the text data. The main work of this paper is as follows: 1. Firstly, this paper discusses the significance of text clustering algorithm and the research status at home and abroad. This paper analyzes the shortcomings of the traditional text clustering algorithms and studies the basic principle of convolution neural network. Based on the transformation of word vector and the basic principle of word2vec tool, a text feature extraction method based on convolutional neural network is designed, and a convolutional neural network model is built. The parameters of the convolutional neural network are selected. The effectiveness of the text feature extraction method based on convolutional neural network is verified by experiments. Finally, an improved KSDM clustering algorithm based on k-means is designed, which is based on the traditional K-means algorithm. A new outlier detection algorithm and a new clustering center selection algorithm are proposed to overcome its shortcomings. The experimental results show that the KSDM algorithm is effective. 5. On the basis of theoretical research, a new word2vec tool is proposed. Combining convolutional neural network with KSDM clustering algorithm, a document analysis framework based on convolutional neural network is proposed. Then the obtained word vector is stored in the vector matrix and input into the pre-trained convolution neural network to extract the text features. Finally, the obtained features are input into the KSDM clustering algorithm for clustering. The experimental results show that the proposed algorithm can effectively improve the accuracy of the existing text clustering algorithms and has a high scalability and flexibility.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.1;TP183

【相似文献】