Aspect-level Sentiment Analysis Based On Deep Learning

With the maturity of Internet technology and the unprecedented development of self-media,netizens can easily express their views and opinions on events through various online platforms,thereby generating a lot of emotional text content.Analyzing and processing these texts and digging out potential content has great social significance and value in applications such as market research,user behavior and Internet public opinion.Generally speaking,for traditional sentiment analysis,it is usually divided into dictionary-based methods and machine learning-based methods.Among them,dictionary-based methods are needed to create a set of emotional word sets in various fields.The extent depends on the quality of the word collection.As new network words continue to appear on the Internet,the use of dictionary-based methods is greatly reduced.The method based on traditional machine learning requires manual selection of features,which is not only time-consuming and labor-intensive,but also the difference in feature selection will also affect emotional judgment and weaken the generalization ability of the model.With the emergence and development of deep learning,the method based on deep learning has long become one of the topics that need to be studied.If it is compared with traditional machine learning methods,the emotion analysis method based on deep learning can automatically extract text Characteristics,and this method has relatively better performance and good expressiveness.Therefore,this paper specifically focuses on deep learning-based sentiment analysis for systematic research.The following is the specific research content.(1)First of all,this thesis comprehensively summarizes the technical research related to sentiment analysis.As far as text representation is concerned,the traditional one-hot representation method is replaced by a neural network-based representation method because it cannot represent the semantic information contained in a sentence;On the one hand,traditional machine learning methods mainly include K nearest neighbors,naive Bayes,and support vector machines.The sentimental features of text must be extracted from a manually labeled corpus;but sentiment analysis methods based on deep learning can usually be divided into three Categories,and all use neural networks to automatically extract features.(2)This article mainly studies aspect-level sentiment analysis,which is a fine-grained sentiment classification task that performs sentiment analysis for specific targets appearing in text sentences.Therefore,we have studied and implemented several types based on different long and short-term memory networks(LSTM).Text sentiment classification methods and related comparisons have laid a good foundation for follow-up research.(3)Aiming at the sentiment analysis at the target level,this article uses Bi-GRU for sentiment analysis,and incorporates location information into the text word vector,combined with the cross-attention mechanism to enhance the interactive relationship between text word vectors,Thereby improving the accuracy of emotion classification.By designing and implementing relevant comparative experiments on the Sem Eval2014 data set,the effectiveness of the proposed model is verified.Finally,the full text is summarized and the future direction is prospected: For different specific targets appearing in text sentences,multiple attention mechanisms are used to perform aspect-level sentiment analysis,and finally combined to judge the overall sentiment trend of the text sentence.