Can I read DependencyMerge Guided Latent Tree Structure for Aspect-Based Sentiment Analysis on EtoBox?

DependencyMerge Guided Latent Tree Structure for Aspect-Based Sentiment Analysis by Chuan Fang is a scholarly article available to read on EtoBox.

What is DependencyMerge Guided Latent Tree Structure for Aspect-Based Sentiment Analysis about?

Recent studies on aspect-based sentiment analysis (ABSA) demonstrated the efficiency of using task-specific latent trees to model the interaction between aspects and opinion words. However, as information propagates mostly between neighbors, a tree inducer module often produces a random structure at early iterations, which makes the performance unstable. Therefore, this paper uses multiple dependency trees from off-the-shelf parsers as the structured guidance for a latent tree to conquer the unstable problem. A simple yet effective dependency ensemble technique, DependencyMerge, is proposed to facilitate different parsers. With the rich prior knowledge, the latent tree-inducing process is stable and the latent tree is more task-specific. Experiments on five ABSA datasets show that the proposed model has better performance compared to the recent strong baselines.

Author

Chuan Fang

Publisher

IEEE

Published

2022

Language

EN