Research On Fine-grained Sentiment Analysis Based On Deep Learning

Fine-grained sentiment analysis is to mine the sentiment tendency of a specific attribute in the text,so it has important applications in the fields of e-commerce reviews,social networks and public opinion analysis.If you implement a fine-grained sentiment analysis of e-commerce reviews,you can understand the customer's sentiment tendency toward a certain attribute of the product,and adjust the product and sales strategy accordingly according to the customer's attitude.And the basic task of fine-grained sentiment analysis plays an extremely important role in tasks such as system recommendation.Therefore,the study of fine-grained sentiment analysis on text has great significance.The research results of this article are as follows:1 ? A model of Bi GRU fine-grained sentiment analysis based on multi-layer attention mechanism is proposed.In order to obtain more potential information of the text sequence,the feature vector of the pre-trained model trained by the Word2 Vec tool is used as the input of the model,and the double-layer Bi GRU model is used to encode the input text vector.The two-level attentional mechanism is used to automatically assign weight to important local information in the text sequence to improve the utilization of the effective information in the sentence.At the same time,it can reduce the effect of invalid information and noise data errors in the text.Through experimental verification,the comparative analysis with other relevant deep learning models shows that the multi-layer attention mechanism can be used to mine more effective information in sentences.2?Since the convolutional neural network can obtain the local features of the text,a new fine-grained emotion analysis model is developed based on the multi-layer attention mechanism Bi GRU and the convolutional neural network model.Experiments show that the performance of the model can be improved effectively on F1 values of each dimension.3?In natural language processing,the coding layer is usually constructed by a bidirectional recurrent neural network model.In order to make the model more rich and latent in the codinglayer,this paper builds a two-layer bidirectional recurrent neural network coding layer through residual thinking.That is,the coding layer of the model in this paper is built by two bidirectional GRU models,in which the bidirectional GRU input of the second layer of the coding layer is composed of the output of the upper layer and the text input.This kind of residual structure makes the coding layer fully express the semantics?Afterwards,two layers of bidirectional GRU are used in the decoding section to obtain global information,a multi-layer attention mechanism and a CNN with multi-granular convolution kernel to obtain important local features.The combine of the two achieves full decoding,and finally be categorized by full link layer.Through experimental verification,the F1 value of the model in this paper can reach 70.01%,which is greatly improved in terms of F1 and accuracy compared to the same type of deep neural network model.