Research On Short Text Sentiment Classification Model Based On Deep Learning

The sentiment classification in natural language processing is used to mine the emotional information in user comments.The traditional sentiment classification methods have problems such as large data dimension,sparse data,insufficient feature extraction,etc.In order to solve such problems,this paper uses deep learning methods from the model.Fusion and feature fusion are two angles to extract features from user comments.Firstly,it analyzes the research status of sentiment classification,summarizes the problems of text-based emotion classification algorithm based on machine learning,and the corresponding improvement methods made by researchers at home and abroad.At the same time,it analyzes the shortcomings of text-based emotion classification algorithm based on deep learning.And propose corresponding improvement methods..Secondly,aiming at the problems of convolutional neural networks and recurrent neural networks,it is proposed to fuse the two models and make full use of the advantages of different networks to weaken the shortcomings of single networks.For the parameter sharing mechanism of convolutional neural networks,the problem of key features cannot be emphasized.The attention mechanism is introduced to assign different weights to each word in the comment.Considering that user comments are serialized data with context information,use two-way long The short-term memory network extracts the global features in the comments and thereby improves the accuracy of the user's sentiment classification.Thirdly,when extracting features from the largest pooling layer of convolutional neural networks,only the problem that the largest eigenvalues will lose a lot of information is extracted.This paper proposes to combine the maximum eigenvalues with background information as the features extracted by convolutional neural networks.At the same time,the two-way gated neural network is used and a local attention mechanism is introduced to extract the global features weighted by the local features in the user comments.The features extracted from the two networks are merged as the final features of the user comments and the user's emotions are classified.Finally,the two models mentioned in this paper are experimentally verified in real data sets.The results show that the two models can effectively improve the accuracy of text sentiment classification.