Researchers from the NOWJ team have detailed their multi-stage pipeline approach for the COLIEE 2026 competition, which focuses on legal retrieval and reasoning tasks. Their methodology involves a combination of filtering, dense retrieval with various embedding models, and reranking techniques for legal case retrieval. For entailment tasks, they employed LLM-based verification and retrieval-augmented generation frameworks. The team also introduced dynamic routing for query difficulty classification and utilized hierarchical transformers with CRF layers for legal judgment prediction. AI
IMPACT This research showcases advanced techniques for applying LLMs and other AI models to complex legal reasoning and retrieval tasks.
RANK_REASON The cluster contains a research paper detailing methodologies for a competition. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
