Abstract:

Convolution neural network method based on word vector has achieved very good results in sentiment analysis. However, the semantic word vector learned from the context ignores the word sentiment polarity, and the traditional convolution neural network model does not take into account the sentence structural information. Aiming at these shortcomings, this paper proposes a dynamic multi-pooling convolution neural network sentiment analysis model based on sentiment vector. The skip-gram model and sentiment dictionary are used to train the sentiment word vector, and the dynamic multi-pooling strategy is used to segment the sentence and retain the maximum number of feature values. Compared with machine learning model and traditional convolution neural network model, the experimental results show that the accuracy of dynamic multi-pooling convolution neural network model in sentiment analysis is significantly improved.

Key words: sentiment analysis, deep learning, sentiment word vector, convolution neural network, dynamic multi-pooling

摘要：

基于词向量的卷积神经网络方法在情感分析研究中取得了很好的效果。然而，该方法从上下文学习的语义词向量忽略了词语本身的情感极性，传统的卷积神经网络模型未考虑句子的结构信息。针对这两方面的不足，提出了一种基于情感词向量的动态多池卷积神经网络情感分析模型，利用skip-gram模型和情感词典来训练情感词向量，并采用动态多池的策略来分割句子，保留了多个最大特征值。实验结果表明，动态多池卷积神经网络模型在情感分析任务上的准确率较机器学习模型和传统卷积神经网络模型都有显著提升。

关键词: 情感分析, 深度学习, 情感词向量, 卷积神经网络, 动态多池