Dependency Tree Distance for Aspect Level Sentiment Analysis

When dealing with complex statements containing multiple aspects in the existing model, most of them ignore the dependencies between context and aspect words, this makes it difficult for the model to learn local information about aspects. To solve this problem, an aspect-level emotion analysis model based on dependency tree distance is proposed. It can calculate the dependency tree distance between aspect words and other context words according to their position in the syntactic dependency tree structure, then the sentence is divided into local information based on dependency tree distance. In the design of the model structure, the fusion learning method is used. The multi-head attention mechanism is used to fuse the local information, global information and aspect information, and the hidden state containing rich sentence expression is obtained as the classification basis. To verify the effectiveness of the model, experiments are carried out on three benchmark datasets, including Restaurant and Laptop datasets of SemEval 2014, and Twitter datasets. In addition, a comparative experiment is designed for the two processing methods and the multi-group dependency tree distance threshold, and the processing method and the optimal threshold suitable for different datasets are selected.