Temporal Semantic Attention Network for Aspect-Based Sentiment Analysis

Bin Yang, Xinyang Tong, Ying Xing, Qi Shen, Huiying Zhao, Zhipu Xie

Abstract: Aspect-based sentiment analysis aims to predict the polarity of sentiment towards a specific aspect in the context. In this paper, we propose the Temporal Semantic Attention Network (TSAN) model for ABSA tasks, which comprising a Global Semantic Feature Network for feature extraction and an Interact Dual Attention module to capture the dependencies of text-target interaction. Experiments on four ABSA benchmark datasets validates the effectiveness of our modules in extracting aspect-level sentiment features.