The article discusses the development of the Extensible Multi-Granularity Fusion (EMGF) network for Aspect-Based Sentiment Analysis (ABSA). The network integrates information from various sources such as syntactic, attention semantic, and external knowledge graphs to enhance semantic features in ABSA models. It aims to efficiently harness the combined potential of each granularity feature and their synergistic interactions. The paper emphasizes the use of Graph Neural Networks (GNNs) on dependency and constituent trees for syntactic analysis. The experimental findings on SemEval 2014 and Twitter datasets confirm the superiority of EMGF over existing ABSA methods.
Publication date: 12 Feb 2024
Project Page: https://anonymous.4open.science/r/EMGF-E7A6
Paper: https://arxiv.org/pdf/2402.07787
