A Dependency Syntactic Knowledge Augmented Interactive Architecture for End-to-End Aspect-based Sentiment Analysis

The end-to-end aspect-based sentiment analysis (ABSA) task remains to be a long-standing challenge, which aims extract the aspect term and then identify its orientation. In previous approaches, explicit syntactic structure of sentence, reflects syntax properties natural language hence is intuitively crucial for extraction recognition, insufficiently modeled. this paper, we thus propose novel dependency knowledge augmented interactive architecture with multi-task learning ABSA. This model capable fully exploiting (dependency relations types) by leveraging well-designed Dependency Relation Embedded Graph Convolutional Network. Additionally, design simple yet effective message-passing mechanism ensure that our learns from multiple related tasks in framework. Extensive experimental results on three benchmark datasets demonstrate effectiveness approach, significantly outperforms existing state-of-the-art method. Besides, achieve further improvements using BERT as an additional feature extractor.