BERTilda Framework for Topic Split and Merge Detection
August 20, 2026
BERTilda tracks longitudinal text streams by constructing temporal topic graphs using embedding-based models and bidirectional coverage signals. The system uses graph-based rules to explicitly identify topic births, deaths, splits, and merges across time windows.
HOW THIS AFFECTS YOU
●
builderYou can build more sophisticated trend-tracking dashboards that show how themes evolve or fragment over time.
●
designerThis enables new ways to visualize the lifecycle and structural reorganization of information in data products.