The Research On Aspect-level Sentiment Analysis Based On Deep Learning

In recent years,due to the widespread popularity of smart devices and the Internet worldwide,the use of social media,forums and e-commerce sites has been widely improved.Opinion sharing on the Internet has become the norm of modern society,and feedback from users has also had a great impact on the final product.Therefore,sentiment analysis of these comments is crucial.The aspect-based sentiment analysis method will be the best solution to efficiently analyze these comments.It identifies the sentiment of each attribute at the fine-grained level and assists decision-making more effectively than the early sentiment analysis model.Deep learning provides multiple architectural options for modeling sentiment analysis tasks,and has surpassed other machine learning methods to become a cutting-edge method for performing sentiment analysis tasks.Therefore,the aspect-level sentiment analysis model based on deep learning shows better results in analyzing comments.At present,although the graph convolutional network model with attention mechanism has played a huge role in solving aspect-based sentiment analysis tasks,there are also many challenges,such as paying too much attention to aspect features and ignoring the overall semantics,so that the connection between aspect and context semantics cannot be accurately established.In recent years,with the dominance of images,audio and video in social media,graphic content has become an important part of social media.This multimodal information combines text and image,which is a new language model.It is necessary to analyze its emotional polarity.Therefore,how to effectively analyze the information expressed by images and text content in comments has become an urgent problem to be solved in multimodal sentiment analysis tasks.In view of the above problems,this paper mainly carried out the following research:1.Aiming at the problem that the graph convolutional network with attention mechanism cannot effectively balance the weight between aspect features and overall semantics in the aspect-level sentiment classification task,and pay more attention to aspect features while ignoring the overall semantics,a dual-weight graph convolutional network model based on aspect-level sentiment analysis is proposed.The dot product weighting method is used to calculate the alignment scores of aspect and context features,and integrated through an adjustable coefficient.The context features combined with the final weight not only highlight the aspect-related features,but also effectively retain useful information about the overall semantics;finally,the convolutional neural network algorithm with position embedding attention is used to extract the relevant polarity of aspect position and its emotion.The experimental results on three datasets show that the model can effectively improve the effect of aspect-level sentiment classification and achieve better prediction of target aspects and their emotions.2.In the task of dealing with multi-modal aspect-level sentiment analysis,it is easy to be insensitive to the correlation recognition of aspect words and opinion words,and the contribution of image information to word representation is not fully obtained.To solve these problems,a multi-view interactive learning network model is proposed,which extracts features from context and syntax respectively,and models the relationship between text,image and aspect.At the same time,the interactive representation of different modalities is fused to dynamically obtain the contribution of image information to each word in the text.Finally,the full connection layer and Softmax layer are used to classify text emotions.The experimental results on two datasets show that the model can effectively enhance the effect of multi-modal aspect-level sentiment classification.