Fine-grained Sentiment Analysis Based On Neural Network And Attention Mechanism

With the development of information technology,the internet platform has become an important channel for people to express their opinions and exchange information.When users use Weibo and e-commerce platforms,they will comment on certain times or products,which generates a large amount of text information containing users' emotions.If we can make use of these data and analyze the potential sentiment information of users,we can provide decision support for consumers and businesses,and even help government departments to formulate policies and correctly guide the dissemination of public opinions.The traditional sentence-level sentiment analysis usually gives an overall sentiment judgment result to the sentence,and cannot make a judgment for a certain attribute/aspect(Some papers also call it target)of the commodity.In today's diversified development of commodity information,it can no longer meet the actual needs,so the aspect-level fine-grained sentiment analysis emerges as the times require,it can analyze user data from multiple dimensions.Such as the phrase “this phone's battery is very durable but a little slow to process.” From the “battery” point of view the phone performed well while from the “processing speed” point of view the phone was somewhat unsatisfactory.In recent years,with the rapid development of deep learning,fine-grained sentiment analysis work is also booming.Therefore,in this paper,the neural network and attention mechanism in deep learning are used as the starting point to study the fine-grained sentiment analysis:(1)In this paper,a Context-Retention Transformation mechanism(CRT)is proposed to combine two classical neural network models of convolutional neural network(CNN)and long short-term memory network(LSTM),so that the model can extract local features and Serialization feature extraction capability at the same time.The fusion of local feature information and serialized feature information can significantly improve the accuracy of fine-grained emotion analysis task.(2)To solve the problem that the traditional attention mechanism always models aspects and contexts separately,and can't use emotional information accurately,this paper also presents a feature-enhanced attention mechanism(FEA),which can make full use of the interactive information between aspects and context to assist emotional judgment.