Stage-wise Fine-tuning for Graph-to-Text Generation

TL;DR

This paper introduces a two-step fine-tuning approach with a novel tree-level embedding method to enhance graph-to-text generation by better utilizing graph structure information, significantly improving performance on benchmark datasets.

Contribution

It proposes a structured graph-to-text model with a two-step fine-tuning process and a new tree-level embedding technique to better encode graph structures in text generation.

Findings

Significant improvement in text generation metrics on WebNLG 2017 dataset.

Effective utilization of graph structure information through tree-level embeddings.

Enhanced performance over traditional token and position embeddings.

Abstract

Graph-to-text generation has benefited from pre-trained language models (PLMs) in achieving better performance than structured graph encoders. However, they fail to fully utilize the structure information of the input graph. In this paper, we aim to further improve the performance of the pre-trained language model by proposing a structured graph-to-text model with a two-step fine-tuning mechanism which first fine-tunes the model on Wikipedia before adapting to the graph-to-text generation. In addition to using the traditional token and position embeddings to encode the knowledge graph (KG), we propose a novel tree-level embedding method to capture the inter-dependency structures of the input graph. This new approach has significantly improved the performance of all text generation metrics for the English WebNLG 2017 dataset.