A dual-memory attention-based aspect-level emotion classification model and method

Abstract

The invention discloses an aspect-level emotion classification model and method based on dual-memory attention, and belongs to the technical field of text emotion classification. The model of the present invention mainly includes three modules: an encoder composed of a standard GRU cyclic neural network, a GRU cyclic neural network decoder introducing a feedforward neural network attention layer, and a Softmax classifier. The model regards the input sentence as a sequence, and based on the attention of the aspect-level word position in the sentence, two memory modules are respectively constructed from the original text sequence and the hidden layer state of the encoder, and randomly initialized through the feedforward neural network attention layer. Fine-tune the attention distribution of the sentence to capture the important emotional features in the sentence, and build an encoder-decoder classification model based on the learning ability of the GRU recurrent neural network for the sequence to achieve the aspect-level emotion classification ability. The invention can significantly improve the robustness of text sentiment classification and increase the classification accuracy rate.

Description

technical field [0001] The invention belongs to the technical field of text emotion classification, in particular to the technical field of natural language processing, and specifically relates to an aspect-level emotion classification model and method based on a dual-memory attention mechanism and an encoder-decoder structure. Background technique [0002] Sentiment analysis, also known as opinion mining, is a research field that analyzes people's subjective feelings such as opinions, emotions, evaluations, opinions and attitudes towards physical objects such as products, services, organizations, individuals, events, topics and their attributes. Aspect-level sentiment analysis is to analyze the emotional tendency (positive, negative or Neutral), which is a subdivision task of sentiment analysis and one of the fundamental concerns in this field. [0003] Traditional feature representation methods include One-hot, N-Gram, and some effective features designed by domain expert...

AI Technical Summary

Problems solved by technology

[0004] Among the existing solutions, the cyclic neural network model based on the attention mechanism and the multi-layer model based on the attention mechanism perform better. The reason for the better performance of the former is that with the help of the feature abstraction mechanism of the deep learning model, more accurate attention distribution, and the latter uses the attention captured by the previous layer to help the next layer to calculate a more accurate attention distribution. Very important word-level or phrase-level emotional features

Images