Research On Aspect-based Sentiment Analysis Based On Deep Learning

Nowadays,e-commerce and social media have become completely popular among the general public,which generates a large amount of textual data,often with users’ sentiment expressions,and studying the sentiment tendency of textual data can provide business or service providers with user feedback references,which brings considerable business value and research significance.Aspect-based Sentiment Analysis(ABSA)is a fine-grained sentiment analysis task that aims to identify the sentiment polarity(e.g.,positive,negative,or neutral,etc.)of one or more given aspect items in a sentence.Although existing aspect-based sentiment analysis methods have achieved excellent performance,there is still room for improvement.In this paper,we propose two graph convolutional neural network based models to address the problems of existing models,and the main work consists of the following two points:(1)Existing aspect-based sentiment analysis models based on graph neural networks usually directly construct sentence dependency trees to represent the dependency information between context and aspect,while ignoring the utilization of word sentiment knowledge information.Second,existing classical aspect-based sentiment analysis models typically split aspect items in a sentence to discriminate their sentiment polarity separately,while neglecting to exploit the dependency information between different aspects.To address the above problems,this paper proposes an aspect-based sentiment analysis model(DGCN-SKIA)based on bipartite graphical convolutional networks to learn sentiment knowledge and inter-aspect dependencies.Specifically,the model firstly acquires intra-contextual dependency information and contextual and aspectual item interaction dependency information based on selfattention mechanism and interaction attention mechanism.Second,Sentic Net sentiment dictionary is used to add sentiment knowledge information to words,and the model builds aspect interaction graph based on the positional relations between aspect items.Finally,the model uses the graph convolutional network to extract features and separately sum the two types of information.In this paper,we experimentally validate the proposed model on four datasets in the restaurant and laptop domains,where the model outperforms the best benchmark model by an average of 1.96% in terms of accuracy and F1 values,and achieves a maximum accuracy of 91.4% on the Rest16 dataset.(2)The summation in fusing information in the DGCN-SKIA model proposed in this paper belongs to a simple static fusion operation,which leads to the model not being able to fuse emotional knowledge information and inter-aspect dependency information adequately.To address the above problems,this paper improves the DGCN-SKIA model and proposes an aspect-based sentiment analysis model(GMFSKIA)based on the gating mechanism to fuse sentiment knowledge information and aspect interaction information.The proposed model first improves the sentiment knowledge graph construction method to model sentiment relations of words in the form of sentiment weights.Second,the model uses aspect-related attention mechanism to model the inter-aspect dependencies.In addition,the model designs an information gate to dynamically fuse sentiment knowledge information and inter-aspect relationship information.Finally,the model uses a masking mechanism and an attention mechanism to obtain aspect-specific representations and perform classification.The proposed model improves DGCN-SKIA by up to 2.01% and achieves 91.56% accuracy on the Rest16 dataset.