A Variant of Recurrent Entity Networks for Targeted Aspect-Based Sentiment Analysis

Abstract

Deep neural network models have achieved promising results on targeted aspect-based sentiment analysis. However, previous models did not effectively match long-distance fine-grained sentiment polarity with the associated target and aspects, and the interdependence among the specific target, corresponding aspects, and the context is always ignored. This work proposes a novel recurrent entity memory network that employs word-level information and sentence-level hidden memory to entity state tracking. In addition, the entity state is utilized to fine-tune the target embedding and aspect embedding. The experimental results showed that the proposed model outperformed previous models.

Topics & Concepts