BERT short text sentiment analysis method for improving training mode

A BERT short text sentiment analysis method for improving a training mode comprises the steps that a short text sentiment analysis model is constructed, and the short text sentiment analysis model comprises an input layer, a semantic feature extraction layer, a pooling layer, a full connection layer and a classification output layer; a data set is collected and preprocessed; encoding in the input layer to obtain word vector representation of the input text; adding disturbance in the word vector to obtain an adversarial sample; the semantic feature extraction layer performs semantic feature extraction on the adversarial sample based on a BERT model, and outputs a feature vector; after a pooling layer and a full connection layer are adopted, Softmax is utilized to carry out normalization processing so as to obtain a final emotion polarity classification result; the short text sentiment analysis model is trained in an adversarial training mode, the problems that sentiment is wrongly divided due to polysemy of Ch