CNN-based emotion analysis method, apparatus and device, and storage medium

Abstract

The invention discloses a CNN-based emotion analysis method, apparatus and device, and a storage medium. In the scheme, a convolutional neural network CNN is used as a basic architecture, and a targettext to be analyzed is input into the CNN to obtain an analysis result. By using the CNN, parallel calculation can be well performed in the training process. A convolutional and gating mechanism layer in the CNN contains a GTRU unit with embedded aspect information, and emotion characteristics can be better controlled through a gating mechanism with the help of given aspect information through the GTRU unit. Furthermore, in a loss function of the CNN full connection layer, a regular item of at least one language rule is contained, and the regular item is formed by modeling through a linguistic resource and contributes to emotion polarity recognition. According to the scheme, the interpretability of a CNN model can be improved through regular items of language rules, and the performance ofthe CNN model is improved.

Description

technical field [0001] The present invention relates to the technical field of sentiment analysis, and more specifically, relates to a CNN-based aspect-level sentiment analysis method, device, equipment, and computer-readable storage medium. Background technique [0002] With the rapid growth of text review data, sentiment analysis is indispensable in modern social networks. It provides the foundation for downstream applications such as intelligent environment, customer analysis system, and text data security protection. Aspect-Based Sentiment Analysis (ABSA) This is a challenging core subtask in sentiment analysis. The purpose of ABSA is not to predict the overall sentiment polarity of a sentence, but to judge the sentiment polarity (positive, negative, or neutral) of various specific aspects in a sentence. For example, in the sentence "The environment is very beautiful, but the food is dreadful", two different aspects are mentioned: "environment (environment), food (food...

AI Technical Summary

Problems solved by technology

However, although these methods are effective, they still have shortcomings, such as: neural networks rely heavily on a large amount of data, and this purely data-driven learning will have unexplainable results

Images