分段卷积神经网络在文本情感分析中的应用

发布时间：2018-08-12 13:35

【摘要】：文本情感分析是当前网络舆情分析、产品评价、数据挖掘等领域的重要任务。由于当前网络数据的急剧增长,依靠人工设计特征或者传统的自然语言处理语法分析工具等进行分析,不但准确率不高而且费时费力。而传统的卷积神经网络模型均未考虑句子的结构信息,并且在训练时很容易发生过拟合。针对这两方面的不足,使用基于深度学习的卷积神经网络模型分析文本的情感倾向,采用分段池化的策略将句子结构考虑进来,分段提取句子不同结构的主要特征;并且引入Dropout算法以避免模型的过拟合和提升泛化能力。实验结果表明,分段池化策略和Dropout算法均有助于提升模型的性能,所提方法在中文酒店评价数据集上达到了91%的分类准确率,在斯坦福英文情感树库数据集五分类任务上达到了45.9%的准确率,较基线模型都有显著的提升。
[Abstract]:Text emotion analysis is an important task in the field of network public opinion analysis, product evaluation, data mining and so on. Because of the rapid growth of network data, it is not only accurate but also time-consuming and laborious to analyze by means of artificial design features or traditional natural language processing grammar analysis tools. However, the traditional convolution neural network models do not take into account the structure information of sentences, and they are easy to be fitted in training. In view of these two shortcomings, we use the convolution neural network model based on deep learning to analyze the emotional tendency of the text, and adopt the strategy of segmentation pool to consider the sentence structure and extract the main features of the different structure of the sentence. Dropout algorithm is introduced to avoid model overfitting and improve generalization ability. The experimental results show that both the piecewise pool strategy and the Dropout algorithm can improve the performance of the model, and the proposed method achieves a classification accuracy of 91% on the Chinese hotel evaluation data set. The accuracy of the five classification tasks in the Stanford English emotion tree database is 45.9%, which is significantly higher than that of the baseline model.
【作者单位】：北京交通大学经济管理学院;
【分类号】：TP391.1

【参考文献】