Aspect-level multi-modal sentiment analysis method based on double channels and attention mechanism

The invention relates to an aspect-level multi-mode sentiment analysis method based on two channels and an attention mechanism, which is characterized in that on the basis of a neural network, sentiment information contained in image features is extracted in a multi-scale manner by combining aspect word features and text features with the attention mechanism, and a GCN network is introduced into an aspect-level multi-mode sentiment analysis task, so that the sentiment analysis efficiency is improved. And the feature extraction and interactive fusion capabilities of the model are greatly improved. According to the method, aspect words, text features and image features are extracted in a feature extraction layer by adopting a pre-training encoder, and final aspect word feature and sentence feature representation is obtained after bidirectional fusion of aspect word and sentence features in an attention mechanism layer. An image feature extraction network is established for image features through a channel atten