融合双通道语义特征的情感分析研究

Research on Sentiment Analysis Fused with Dual Channel Semantic Features

现有大多数深度学习模型结构单一,通常会降低文本语义特征提取能力。为此,提出一种融合双通道语义特征(FDSF)的情感分析研究模型。首先,采用BERT预训练语言模型获取文本的动态特征向量表示。然后,将BiG-RU-Attention通道提取的全局语义信息经注意力动态权重调整后,与CNN通道提取的局部语义信息进行特征向量融合。最后,将融合特征经过全连接层与Softmax函数,输出最终情感极性。在ChineseNLPcorpus的online_shopping_10_cats、中科院谭松波学者整理的数据集上与现有主流情感分析方法进行比较实验,结果表明,FDSF模型在F1值与准确率方面均最优,证明了该模型在情感分析任务中的有效性和可行性。

Most existing deep learning models have a single structure,which usually reduces the ability to extract text semantic features.To this end,a sentiment analysis research model integrating dual channel semantic features(FDSF)is proposed.Firstly,the BERT pre trained language model is used to obtain the dynamic feature vector representation of the text.Then,the global semantic information extracted by the BiGRU Attention channel is adjusted by attention dynamic weights,and fused with the local semantic information extracted by the CNN chan-nel for feature vectors.Finally,the fused features are processed through a fully connected layer and Softmax function to output the final emo-tional polarity.Experiment on online_shopping_10_cats of ChineseNLPcorpus,and dataset compiled by scholars Tan Songbo from the Chinese Academy of Sciences,compared with existing mainstream sentiment analysis methods,the FDSF model has the best F1 value and accuracy,proving its effectiveness and feasibility in sentiment analysis tasks.