发表机构
Faculty of Sciences, University of Porto; INESCTEC（波尔图大学理学院; INESCTEC）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
BERTilda是基于嵌入的可解释时序主题框架，通过构建含语义相似度与双向覆盖信号的主题图检测主题分裂/合并等转换，在政治语料库上验证后其生命周期标签标注同意率优于基线方法。
AI 中文摘要
纵向文本流不仅存在主题的诞生与消亡，还存在主题分裂为子主题或合并为更广泛叙事的离散结构重组。许多动态主题模型强调平滑漂移，而快照主题模型（按时间窗口独立拟合）未明确时间对应关系。本文提出BERTilda，这是一种可解释框架，它在每个窗口中独立发现主题（使用基于嵌入的主题模型），然后构建连接相邻窗口主题的时序主题图。该图由两个互补信号支持：（i）主题表示之间的语义相似度；（ii）双向覆盖信号，通过跨窗口推文-主题归因估计文档流出（主题去向）和流入（主题来源）。基于图的规则标记延续、分裂、合并、消失和不明确的转换。我们在政治语料库上评估BERTilda，包括美国国会推文和历史演讲数据集，报告主题质量和时间稳定性诊断结果，并在三位独立标注者标注的黄金标准子集上验证生命周期标签。在标注子集上，BERTilda达到最高87%的多数同意率，在对比方法中获得最高宏平均同意率，相对于仅相似度和仅正向基线，在消失检测方面表现尤为突出。
英文摘要
Longitudinal text streams exhibit topic birth and death, but also discrete structural reorganizations in which themes split into subtopics or merge into broader narratives. Many dynamic topic models emphasize smooth drift, while snapshot topic models (fit independently per time window) leave temporal correspondence underspecified. We present BERTilda, an explainable framework that discovers topics independently in each window (using an embedding-based topic model) and then constructs a temporal topic graph linking topics across adjacent windows. Links are supported by two complementary signals: (i) semantic similarity between topic representations and (ii) a bidirectional coverage signal that estimates document outflow (where a topic goes) and inflow (where a topic comes from) via cross-window tweet-to-topic attribution. Graph-based rules label continuations, splits, merges, disappearances, and unclear transitions. We evaluate BERTilda on political corpora, including U.S. congressional tweets and historical speech datasets, report topic-quality and temporal-stability diagnostics, and validate lifecycle labels on a gold-standard subset annotated by three independent annotators. On the annotated subset, BERTilda reaches majority agreement rates up to 87% and attains the highest macro-average agreement across the compared methods, with particularly strong disappearance detection relative to similarity-only and forward-only baselines.
Comments16 pages, 2 figures, 7 tables, with 4-page supplementary material. Accepted at ECML PKDD 2026 (Naples, 7-11 September 2026). Authors' accepted version; the revised version of record will appear in the proceedings (Springer, Lecture Notes in Computer Science)
