Research On Fine-grained Sentiment Analysis Based On Gated Convolutional Network

As an important task in the field of natural language processing,fine-grained emotional analysis of text aims to extract fine-grained elements such as products,services,events and other fine-grained elements from valuable comments text information on the Internet,and to distinguish the relevant fine-grained elements.At present,fine-grained emotional analysis has been widely used in social network,financial service,public opinion analysis and other aspects,which shows its great value.In recent years,the continuous development of deep learning provides a series of solutions for fine-grained emotional analysis.Although these schemes have achieved good results,the complexity of calculation and the size of the model are greatly increased,which makes the efficiency of emotional analysis low.Especially in the network with massive review data,the low efficiency fine-grained emotional analysis scheme means that the time cost rises sharply.In view of the high efficiency of gate convolution network,this paper proposes an improved fine-grained emotion analysis algorithm based on gate convolution network to solve the efficiency problems existing in the current mainstream methods.The main work of this paper is as follows:(1)This paper studies how to extract fine-grained elements from comment text.Because of the problems of low computational efficiency and poor overall connection extraction ability of the main methods such as recurrent neural network and conditional random field,this paper proposes a text fine-grained element extraction model based on gated convolutional network.In addition,the problem of boundary location error often occurs in the fine-grained element extraction process.This paper attempts to solve the problem by using pointer network to locate the span interval.(2)The current method of text feature representation is to combine the related semantics directly by using concat eigenvector,but the mapping of feature vectors to feature space is not uniform,which leads to the introduction of some impurity information.In this paper,from different perspectives of emotional dictionary,syntactic structure feature,location coding and other aspects of text feature representation,this paper adopts the method of multi-channel text representation vector fusion to extract the semantic information contained in the text more effectively.(3)The research on how to quickly analyze the fine-grained elements extracted from text;most of the current solutions are based on pre-training Bert model or recursive neural network,but these schemes are complex and inefficient,so they can not respond quickly.So combining the text feature representation method in work 2,this paper proposes a fine-grained emotional analysis model of gate interactive dilated convolution network.At the same time,it can improve the accuracy of fine-grained emotional analysis,and significantly reduce the time complexity and size of the model in the training process of the whole model.To verify the effect of the design model,this paper uses the open short text review data semeval2014 as the data set,designs different contrast and ablation experiments,and selects appropriate evaluation indexes to determine the final experimental results.The results show that compared with the main way of Bert pre-training or RNN network,the accuracy and recall rate of the model are improved slightly,and the size and complexity of training time are optimized obviously.