摘要
随着互联网的发展,网上购物成为主流消费方式,随之产生了大量的商品文本数据,需要对商品进行准确而高效的分类。利用机器学习进行文本分类需要进行复杂的人工设计特征和提取特征过程。随着深度学习领域的发展,基于深度学习的文本分类技术效果显著。设计了一个基于长短期记忆网络(LSTM)的中文文本多分类器。首先对数据进行预处理,利用Tokenizer分词技术将文本处理为计算机可理解的词向量传入LSTM网络,并加入Dropout算法以防止过拟合得出最终的分类模型。将该模型与逻辑回归、多项式朴素贝叶斯、线性支持向量机、随机森林模型进行对比发现,基于LSTM的中文文本多分类方法具有较好的效果。
With the development of the Internet,online shopping has become the mainstream,and a large amount of commodity text data has been produced.Accurate and efficient classification of commodities has become a demand.Text classification by machine learning requires a complex process of designing and extracting features manually.With the development of deep learning,text classification technology based on deep learning has achieved remarkable results.Therefore,this paper designs a multi-classifier of Chinese text based on Long Short-Term Memory.First of all,the data is preprocessed,the text is processed into the word vector that can be understood by the computer and sent to the LSTM network by the Tokenizer word segmentation technology,and the Dropout algorithm is added to prevent overfitting to get the final classification model,and the model is compared with logistic regression,naive bayes,support vector machine and random forest.The experimental results show that the LSTM-based multi-classification method has achieved good results.
作者
梁登玉
LIANG Dengyu(School of Computer Science and Technology,Shanghai University of Electric Power,Shanghai 200090,China)
出处
《上海电力大学学报》
CAS
2020年第6期598-602,共5页
Journal of Shanghai University of Electric Power
关键词
文本多分类
深度学习
长短期记忆网络
自然语言处理
multi-classification
deep learning
long short term memory
natural language processing