AMR-based Path Aggregation Graph Network for Aspect-based Sentiment Analysis

Anonymous

Abstract: Aspect-based sentiment analysis (ABSA) is a fine-grained sentiment classification task. Many recent works have used dependency trees to extract the relationship between aspects and contexts and have achieved significant improvements. However, further improvement is limited due to the mismatch between the dependency tree as a syntactic structure and the sentiment classification as a semantic task. To alleviate this gap, we replace the syntactic dependency tree with the semantic structure, Abstract Meaning Representation (AMR) and propose a model called AMR-based Path Aggregation Graph Network (APAGN). Particularly, we design a path aggregation module which collect local information into global information by path to make full use of AMR. APAGN also contains the outer product summary module which transfers the feature from sentence to graph and the relation-enhanced attention mechanism which transfers the feature in the opposite direction. Experimental results on three public datasets demonstrate the effectiveness of APAGN in aspect-based sentiment analysis when compared with baselines.

Paper Type: long