基于门控交互膨胀卷积网络的属性情感分析
首发时间:2021-03-25
摘要:\justifying 属性情感分析作为细粒度的情感分析问题,主要目的是明确特定目标的情感倾向。以往的细粒度情感分析方法通常使用循环神经网络、预训练BERT模型作为特征提取器,在情感分析上取得了不错的结果,但限制于循环神经网络依次处理序列的特性和BERT模型的复杂网络结构,使得训练和运行的效率比较低下,耗时较多。针对该问题,结合门控卷积网络具有较高计算效率的特点,提出融合多通道文本特征和门控交互膨胀卷积网络的模型,将门控卷积网络作为特征编码器,提高模型训练的效率;同时引入多通道文本特征,对上下文进行特征表达以增强与实体属性词相对应的情感要素提取。在SemEval2014数据集上的实验,在餐厅数据集和笔记本电脑数据集上取得了81.96$\%$和75.39$\%$的准确率,同时在计算效率方面相比于DAuM模型和BERT模型分别提升了3倍与22倍。
关键词: 人工智能 属性情感分析 门控交互膨胀卷积网络 多通道文本特征 计算效率
For information in English, please click here
Sentiment Analysis of Aspects Based on Gated Interaction Dilated Convolution Neural Network
Abstract:\justifying Attribute sentiment analysis as a fine-grained sentiment analysis problem, its purpose is to clarify the emotion of specific aspect. For the above problems, most of the previous methods are combined with recurrent neural network and BERT to construct the network model. However, recurrent neural network is difficult to parallelize the model due to their time sequence characteristics. In order to solve this problem, combined with the high computational efficiency of gated convolutional networks, combining multi-channel text features and gated interactive dilated convolutional networks is proposed. The gated convolutional network is used as a feature encoder to improve the efficiency of model training. At the same time, using multi-channel text features enhances the feature extraction effect in the context, and then captures the emotional elements corresponding to the entity attribute words. Experiments on the SemEval2014 data set have achieved a high accuracy rate. In addition, compared with DAuM and BERT models in terms of computational efficiency, it has been improved by 3 times and 22 times.
Keywords: \justifying artificial intelligence aspect sentiment analysis gated interactive dilated convolutional networks multi-channel text features computational efficiency
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于门控交互膨胀卷积网络的属性情感分析
评论
全部评论0/1000