基于距离感知的目标情感分类模型

Distance-Based Model for Target-Level Sentiment Analysis

目标情感分类是是一种细粒度的情感分类任务,针对人工生成特征模型成本高且不能捕捉上下文语义、传统循环神经网络模型训练时间长等问题,设计了一种基于距离感知的目标情感分类模型.通过距离感知窗口对目标词与邻近词之间的距离信息进行建模,结合词嵌入技术,分别对输入文本和距离信息建立向量矩阵,使用卷积神经网络提取特征,将文本语义特征和距离特征结合,输入到分类层进行目标情感分类.最后在SemEval2014笔记本电脑和餐厅两个数据集上进行实验,取得了比基于循环神经网络生成特征的模型和利用外部语法分析器生成特征的模型更好的分类效果,且具有更短的模型训练时间.研究结果对目标情感分类领域的应用具有参考价值.

Target-level sentiment analysis is a fine-grained classification task.Aiming at the problems such as high cost of manually generated feature model,inability to capture context semantics,and long training time of recurrent neural network,a target-level sentiment analysis model based on distance is designed.The distance information between the target word and its neighbors is modeled through the distance perception window.Combined with the word embedding technology,the matrix-vector is established for the input text and distance information,respectively.The features are extracted using the convolutional neural network,and the semantic features of the text are combined with the distance features.Finally,experimental results achieved on a SemEval 2014 dataset(Laptop and Restaurant)show that our approach achieves a significant improvement in the accuracy over the comparison models and has a shorter model training time.The research results have reference value for the applied research of target sentiment classification.