Self-attention mechanism social network text sentiment analysis method based on disturbance improvement

The invention relates to a social network text sentiment analysis method based on a disturbance improvement self-attention mechanism. The method is used for analyzing sentiment expressed by a text in a network. The method comprises the following steps: segmenting sentences in web text data into words by using a word segmentation tool, and converting each word into a word vector by using a word embedding matrix; inputting the word vectors into a pre-training language model (BERT-base) to obtain a hidden layer state (feature representation) of each word; inputting the hidden layer state of the word into a classifier to obtain classification probability distribution of the sentence; performing disturbance improvement on the hidden layer state of each word and the classification probability distribution of the sentences to obtain attention supervision information; secondarily training the pre-training language model by using the attention supervision information; and inputting the word vectors into the improved a