Text sentiment classification algorithm based on convolutional neural network and attention mechanism

Abstract

The invention discloses a text sentiment classification algorithm based on a convolutional neural network and an attention mechanism. The text sentiment classification algorithm comprises the steps of1, establishing the convolutional neural network comprising multiple convolutions and multiple kinds of pooling, and using sentiment classification text for training to obtain a first model; 2, establishing the multi-head point product attention mechanism into which residual connection and nonlinearity are added, and using the sentiment classification text for training to obtain a second model; 3, conducting model fusion on the two models to obtain sentiment classification of the text. Multiple granularity, the convolutions and multiple kinds of pooling are fused into the convolutional neuralnetwork, the residual connection and the nonlinearity are introduced into the attention mechanism, and attention is calculated several times to obtain two text sentiment classification models. Through a Bagging model fusion method, a fusion model is obtained, the text is classified, the advantages that the convolutional neural network can well capture local features and the attention mechanism can well capture global information can be combined, and the more comprehensive text sentiment classification models are obtained.

Description

technical field [0001] The invention relates to the field of text classification of natural language processing, in particular to a text emotion classification algorithm based on a convolutional neural network and an attention mechanism. Background technique [0002] Text classification has various applications, such as sentiment polarity classification, topic classification, etc. For text classification, there are many commonly used methods, such as unsupervised methods based on dictionaries and rules, and supervised methods based on machine learning. The dictionary-based method uses authoritative dictionaries and artificially constructs features according to experience. The accuracy of the model is high, but the recall rate of the model is low due to the low coverage of the dictionary. Supervised methods based on machine learning, modeled using machine learning methods such as SVM, maximum entropy model, naive Bayesian, and KNN. These machine learning methods are relativ...

AI Technical Summary

Problems solved by technology

However, ordinary convolutional neural networks set fixed filters and pooling operation types, and the captured local information has a fixed granularity, is relatively rigid, and lacks diversity.

Images