Core Concepts
Gradient-based explanation methods are enhanced by the Information Bottleneck-based Gradient (IBG) framework for Aspect-based Sentiment Analysis (ABSA), improving model performance and interpretability.
Stats
"Our comprehensive evaluations and tests provide substantial evidence of the effectiveness of the IBG framework."
"The key contributions of this paper are listed as follows."
Quotes
"Gradient-based explanation methods are increasingly used to interpret neural models in natural language processing."
"Our model is model-agnostic, we integrate it with several state-of-the-art baselines."