An Aspect Level Sentiment Analysis Method for Graph Convolutional Networks based on Improved Dependencies

Aspect-level sentiment analysis can be more delicate to grasp the user’s emotional tendency, has a wide range of applications in various fields. It has been proved that the method of combining dependency relationship with graph convolutional neural network can be used to improve the effect of aspect-level sentiment analysis. However, user comments are generally conducted from multiple perspectives, resulting in multiple aspects in sentences, and longer comments usually contain more effective information. This will make parsing sentence dependencies more complicated, and generate a large number of parameters in the convolution process, resulting in noise and other problems. Therefore, it is necessary to choose and choose the information and leave the effective information for sentiment analysis. This paper proposes a graph convolution network joint model to improve the dependency relationship(AIDE-GCN). By improving the dependency tree, this model builds multiple matrices and vectors for effectively screening sentences, so that the weight matrix is constructed with the aspect word as the center. When realizing the sentiment analysis of multi-aspect words, only the target dependence is kept, so that the model magnitude is smaller and the accuracy is guaranteed. Through experiments on public datasets SemEval series, Twitter series and more challenging MAMS series, it is proved that the proposed model has obvious advantages.