Effective Token Graph Modeling using a Novel Labeling Strategy for Structured Sentiment Analysis

Anonymous

Abstract: The state-of-the-art model for structured sentiment analysis casts the task as a dependency parsing problem, which has some limitations: (1) The label proportions for span prediction and span relation prediction are imbalanced; (2) Two nodes in a dependency graph cannot have multiple arcs, which are necessary for this task; (3) The losses of predicting the imbalanced labels are directly applied in the prediction layer, which further exacerbate the imbalance problem. In this work, we propose nichetargeting solutions for this issues. First, we introduce a novel labeling strategy, which contains two sets of token pair labels, namely essential labels and whole labels. The essential label set consists of the minimum labels for this task, which are relatively balanced and applied in the prediction layer. The whole label set includes rich labels to help our model capture various token relations, which is imbalanced but merely applied in the hidden layer to softly influence our model. Moreover, we also propose an effective model to well collaborate with our labeling strategy, which is equipped with the graph attention network to iteratively refine token representations, and the adaptive multi-label classification to dynamically predict multiple relations between token pairs. We perform extensive experiments on 5 benchmark datasets in four languages. Experimental results show that our model outperforms previous SOTA models by a large margin. We believe that our labeling strategy and model can be well extended to other structured prediction tasks.