基于情感融合和多维自注意力机制的微博文本情感分析

Micro-blog sentiment analysis based on emotional fusion and multi-dimensional self-attention mechanism

将词向量融入情感信息以及有效提取文本特征的网络结构是提升情感分析准确率的关键。针对传统词向量没有充分利用微博中表情符号的情感特征,以及提取特征的模型通常基于卷积神经网络(CNN)和循环神经网络(RNN),难以克服局域性和不能并行化的问题,提出一种基于情感融合和多维自注意力机制的微博文本情感分析模型E-DiSAN。该模型将融合表情符号的语义合成向量作为网络的输入层,利用融合位置信息的多维自注意力网络提取高层文本特征训练分类器,实现了文本中词语间依赖关系的建立以及多角度情感语义信息的获取,并通过对比实验证明了该模型的有效性。

Integrating word embedding into emotional information and effectively extracting the network structure of text features are the keys to improve the accuracy of sentiment analysis. The traditional word vector method does not make full use of the emotional features of the emojis in micro-blog, and the models for extracting features are usually based on Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN), which are difficult to overcome locality and cannot be parallelized. To solve the problems, a micro-blog sentiment analysis model based on emotional fusion and multi-dimensional self-attention mechanism called Emotional and multi-Dimentional Self-Attention Network (E-DiSAN) was proposed. A semantic synthesis vector fused with emojis was used as the input layer of the network, and a high-level text features training classifier was extracted by using a multi-dimensional self-attention network fused with location information, which realizes the establishment of the dependence relationship between words in the text and the extraction of multi-angle emotional semantic information. The effectiveness of the model was proved by comparative experiments.