Knowledge-driven Augmentation and Retrieval for Integrative Temporal Adaptation
KARITA is a framework for temporal domain adaptation in specialized multi-label text classification. It addresses the challenge of model performance degradation when models trained on historical data are applied to future data, due to temporal distribution shifts in specialized domains.
KARITA introduces:
- Multi-signal shift detection combining uncertainty-based, feature-based, and ontology-based signals to identify samples affected by temporal drift
- Shift-aware retrieval of semantically similar historical samples to provide targeted support for shifted instances
- Knowledge-driven augmentation using external ontologies (MeSH, EuroVoc, CSO) and LLM-based synonym generation to bridge terminological gaps across time periods
The framework is evaluated across three domains: clinical notes (MIMIC-IV-Notes), legal documents (EUR-Lex), and computer science papers (ArXiv-CS).
Code and data preprocessing scripts will be released upon paper acceptance. Stay tuned!
Coming soon.