CNN-BiGRU模型在中文短文本情感分析的应用

摘要：

【目的/意义】改善传统情感分析方法工作量大，以及研究者采用深度学习方法多数仅致力于提高分析准确

率，往往忽略网络训练速度的问题。【方法/过程】提出将卷积神经网络（CNN）与双向门控循环单元（GRU）相结合的

文本情感分析模型（CNN-BiGRU），通过CNN和双向GRU对文本的局部静态特征以及序列特征进行提取，后接

单向GRU层对其进行进一步降维，最后使用Sigmoid进行情感分类。【结果/结论】通过自建豆瓣影视评论数据集，将

本模型与同复杂度的CNN-BLSTM模型相比，分类准确率和训练速率分别提高了2.52%、41.43%。【创新/局限】提出

CNN-BiGRU网络应用于短文本情感分析，简化特征提取过程，引入上下文语义信息，减少参数提高效率。

Abstract:

【Purpose/significance】To improve the workload of traditional sentiment analysis methods, and most researchers use deep

learning methods only to improve the accuracy of analysis, and often ignore the problem of network training speed.【Method/process】

A text sentiment analysis model (CNN-BiGRU) combining a convolutional neural network (CNN) and a bidirectional gated recurrent

unit (BiGRU) was proposed. The local static features and sequence features of the text were extracted through CNN and BiGRU. Fol⁃

lowed by a one-way GRU layer to further reduce the dimensionality, and finally uses Sigmoid for emotion classification.【Result/con⁃

clusion】Compare this model with a CNN-BLSTM model of the same complexity by using a self-built Douban film and television re⁃

view data set, this model improves the classification accuracy and training rate by 2.52% and 41.43% respectively.【Innovation/limita⁃

tion】CNN-BiGRU network is applied to sentiment analysis of short text, which simplifies the process of feature extraction and intro⁃

duces context semantic information to reduce parameters and improve efficiency.

缪亚林, 姬怡纯, 张顺, 程文芳, 彭二楼.

CNN-BiGRU模型在中文短文本情感分析的应用

[J]. 情报科学, 2021, 39(4): 85-91.

可视化