Emotional target extraction model based on multi-word embedding fusion and attention mechanism

The invention relates to an emotion target extraction model ME-ATT-CRF based on multi-word embedding fusion and an attention mechanism. According to the model, three types of word embedding are adopted for fusion, universal embedding and specific domain embedding are adopted, the condition that word forms can reflect part-of-speech to a certain extent and then affect a labeling result is considered, morphological information of character-level convolutional learning words is added to enrich feature representation, and character-level features are extracted. Under the condition that no extra supervision is used, the model achieves a better effect. In addition, a self-attention mechanism is introduced into a hidden layer of the model, so that the model can automatically learn association and weights among different words in an input text, context semantics are fully understood, and more attention is paid to target words to be extracted. Experimental verification and comparison are carried out on the four data s