Research On Method Of Chinese Text Sentiment Classification Based On Deep Learning

The field of text emotion classification faces the dilemma that there are few Chinese text classifications.At the same time,most of the existing emotion classification tasks are mainly based on emotion polarity classification,and do not perform fine-grained emotion classification for Chinese texts,resulting in the lack of special Chinese emotion classification.Corpus.In terms of feature extraction in the field of Chinese text sentiment classification,there are problems such as difficulty in text feature extraction,difficulty in obtaining contextual connection information,and insufficient local feature extraction.Aiming at the above problems,this paper introduces a deep learning algorithm into the Chinese text sentiment classification task to carry out research,and achieves good classification results.This paper first proposes a technical framework for Chinese text sentiment classification.Because most emotion classification tasks are mainly based on emotion polarity classification,fine-grained emotion classification is not performed on Chinese texts,resulting in the lack of a special Chinese emotion classification corpus,and it is difficult to build a vocabulary suitable for Chinese emotion classification.In response to this problem,the dataset selected in this paper is a fine-grained emotion dataset divided into four categories,and it is visualized and analyzed,a suitable Chinese word segmentation algorithm is selected for word segmentation,and finally a vocabulary suitable for Chinese emotion classification is constructed.Aiming at the difficulties of text feature extraction in the field of Chinese text sentiment classification,difficulty in obtaining contextual connection information,and insufficient local feature extraction,a variety of word vector representation methods were compared,and Word2vec was finally selected to represent the word vector,using the semantic features and Syntax interpretability,and text feature selection is performed on the generated word vectors,and feature vectors suitable for Chinese text sentiment classification tasks are selected.Then,on the basis of the traditional convolutional neural network,combined with the attention mechanism,a multi-core convolutional neural network Chinese text sentiment classification algorithm(AM-CNN)combined with the attention mechanism was designed and implemented.The core of the algorithm is convolution and pooling.Some convolution kernels of different sizes are used to convolute and pool separately and then combine to solve the problem of incomplete information of local text features extracted by single-layer convolution and information loss caused by text features extracted by multi-layer convolution structure.The comparative experimental results show that the multi-kernel convolutional neural network Chinese text sentiment classification algorithm combined with the attention mechanism is effective in the field of Chinese text sentiment classification,and the classification precision and recall rate reach 0.8527 and 0.8491,respectively.Then imitate the RGB three-channel input method of convolutional neural network in the field of image processing,improve on the basis of the multi-core convolutional neural network Chinese text sentiment classification algorithm combined with attention mechanism proposed in this paper,design and implement a dual-channel multi-core volume Convolutional neural network Chinese text sentiment classification algorithm(DM-CNN),the dual-channel feature extraction method of this algorithm combines Word2vec and attention mechanism,so that the convolution part can perform convolution operation according to the attention weight,so that it is difficult to improve the contextual connection information.Get the problem.The experimental results show that the accuracy of the improved dual-channel multi-kernel convolutional neural network Chinese text sentiment classification algorithm reaches 0.8647,which is better than the previous AM-CNN algorithm.Then,aiming at the problem of unbalanced dataset samples,a comparison experiment between the unbalanced dataset and the balanced dataset is designed and implemented on the dual-channel multi-kernel convolutional neural network Chinese text sentiment classification algorithm.The experimental results show that the classification on the balanced dataset is accurate.The rate is 0.8713,which is better than the classification effect on the imbalanced dataset.Finally,for deep learning algorithms,which usually encounter problems such as gradient disappearance and overfitting of classification results,a dropout layer is added to the network structure.At the same time,three optimizers are compared on the DM-CNN model,namely Adam optimization algorithm,stochastic gradient descent method and RAdam optimization algorithm.Through the result analysis,the RAdam optimization algorithm is selected as the optimizer and added to the model to avoid over-fitting and gradient.disappear,etc.