TAGHN：文本注意力图重排序用于自动主题标引
原标题：Reading the Neighborhood: Text-Attentive Graph Reranking for Automated Subject Indexing
AI 摘要
研究者提出 TAGHN，一种面向自动主题标引的异构图重排序器，在检索候选描述符的子图上引入叙词表邻域（上位词、下位词、相关词）进行文档条件化消息传递。在包含 6,198 个描述符的经济学叙词表 STW 上，TAGHN 达到 0.380 宏 F1，优于微调交叉编码器的 0.346 和固定词表 XMLC 分类器的 0.278；微观 F1 上优于重排序模型并与最佳 XMLC 模型接近。定性分析显示图结构能纠正逐标签打分无法解决的错误，例如选择更精确的下位描述符而非宽泛上位词。
正文节选
Reading the Neighborhood: Text-Attentive Graph Reranking for Automated Subject Indexing Abstract. Subject indexing assigns descriptors from a controlled vocabulary to bibliographic records, a task characterized by thousands of labels, few positive examples per label, and a long tail of rarely used descriptors. Existing extreme multi-label classifiers treat the vocabulary as a flat list, ignoring the semantic relations that thesauri encode. We present TAGHN, a heterogeneous graph reranker over a