🤖 AI Summary
This study investigates how to effectively integrate discourse structure—specifically Rhetorical Structure Theory (RST) and coreference graphs—to enhance abstractive text summarization. To this end, the authors present the first RST annotations for the XSum dataset, establishing a new benchmark named XSum-RST, and propose a graph fusion architecture combining Graph Attention Networks (GATs) and Multilayer Perceptrons (MLPs). Experimental results demonstrate that the MLP-based fusion approach outperforms baseline methods on the CNN/DailyMail dataset. Furthermore, evaluations on XSum-RST validate the effectiveness of incorporating graph-structured discourse information and delineate its practical limits, revealing notable differences in how discourse structures influence summary generation across distinct datasets.
📝 Abstract
This study aimed to leverage graph information, particularly Rhetorical Structure Theory (RST) and Co-reference (Coref) graphs, to enhance the performance of our baseline summarization models. Specifically, we experimented with a Graph Attention Network architecture to incorporate graph information. However, this architecture did not enhance the performance. Subsequently, we used a simple Multi-layer Perceptron architecture, which improved the results in our proposed model on our primary dataset, CNN/DM. Additionally, we annotated XSum dataset with RST graph information, establishing a benchmark for future graph-based summarization models. This secondary dataset posed multiple challenges, revealing both the merits and limitations of our models.