Graph neural network dialogue emotion recognition method with noise reduction and error correction

The invention relates to a graph neural network dialogue emotion recognition method with noise reduction and error correction, and belongs to the field of natural language processing. The method comprises the following steps: extracting semantic features of utterances in a dialogue by utilizing a pre-training language model; a designed context screening module is used for evaluating the semantic correlation and the information amount of the context, discarding part of noise contexts and establishing a speech dependency relationship; learning context information from the context by using a relational graph neural network, and generating corresponding emotion features; the semantic features and the emotion features are integrated through the feature error correction module, so that the semantic features and the emotion features can supervise each other, and the purpose of correcting part of errors in the features is achieved; sentiment categories are predicted through a full-connection network, and a cross entr