Research On Aspect-based Emotional Analysis Based On Deep Learning

In recent years,with the rise of the Internet and smart devices,a variety of software such as shopping,social networking,food and travel have been developed.The software bring convenience to people’s life,so in daily life,people will publish emotional critical views or views on the Internet,these views and views are with the subjective emotions of the publishers.These subjective opinions and opinions may lead other users to have a bias against the product or service,and even lead public opinion towards it.Therefore,the emotional analysis of these critical opinions is very important.With the wide application of deep learning technology,emotional analysis tends to be fine-grained.Aspect-based sentiment analysis as a fine-grained sentiment analysis,compared with coarse-grained sentiment analysis,aspect-based sentiment analysis can analyze every aspect word in the sentence,and the analysis of the emotional polarity of aspect words is more accurate,so as achieve a deeper analysis of user emotion.This paper studies the aspect-based sentiment analysis,and puts forward two models,the main contents of which are as follows:(1)Aiming at the problem that semantic information is difficult to extract,and that aspect word information is difficult to associate with context information in aspect-based sentiment analysis tasks,a Fused Dual-Channel Semantic Information model(FDCS)is proposed.The model builds two channels through the BERT pre-training model to obtain semantic information at different levels,one is the global information channel,and the other is the sentence information channel.Semantic attention is used to fuse the semantic information of different levels in the dual channels,and the fused semantic information is re-integrated into the global information and sentence information respectively.Finally,the corresponding feature information is extracted according to the different semantic information of each channel.Experimental results on three benchmark datasets show that this model outperforms other models.(2)In recent years,aspect-based sentiment analysis is basically a single mining of semantic or grammatical information,and there is a lack of correlation between semantic information and grammatical information;In addition,the previous models ignore the joint impact of relative distance and grammatical distance on aspect words,and ignore the positional relationship of words in the dependency syntax tree.Aiming at the above problems,a Multi-Information Learning by Fusing Syntax Trees(MILFST)model is proposed.First,the information of the text sequence is captured by a bidirectional longshort-term memory network.Then update the sequence information according to the tree structure of the dependency syntax tree,and then embed the relative distance and syntax distance position information into the text sequence;The semantic information and syntactic information are learned through convolutional neural networks and graph convolutional networks.The optimization of semantic information and grammatical information is realized through the attention mechanism.Finally,the fused information is input into the Softmax classifier for emotional polarity classification.The experimental results show that the model has different degrees of improvement compared with other models on the five datasets.