📄 中文摘要
该研究提出了一种基于数据驱动的方法，自动构建图形化的文档表示。基于Bugueño和de Melo（2025）的最新研究，利用动态滑动窗口注意力模块，有效捕捉句子之间的局部和中等语义依赖关系，以及文档内部的结构关系。通过在所学习的图上训练的图注意力网络（GATs），在文档分类任务中取得了具有竞争力的结果，同时所需的计算资源低于以往的方法。此外，还对所提出的图构建方法在提取式文档摘要中的应用进行了探索性评估，突显了其潜力及当前的局限性。
从全球到本地：学习上下文感知的图表示用于文档分类和摘要
📄 English Summary
From Global to Local: Learning Context-Aware Graph Representations for Document Classification and Summarization
This research proposes a data-driven method for automatically constructing graph-based document representations. Building on the recent work of Bugueño and de Melo (2025), it leverages a dynamic sliding-window attention module to effectively capture local and mid-range semantic dependencies between sentences, as well as structural relations within documents. Graph Attention Networks (GATs) trained on the learned graphs achieve competitive results in document classification while requiring lower computational resources than previous approaches. An exploratory evaluation of the proposed graph construction method for extractive document summarization is also presented, highlighting both its potential and current limitations.