Atlas · GenAI 2026
Search Re-Ranking
Cross-encoders & re-ranking
conceptPeak: 2024Retrieval TechniquesAI consensus: 3/3
Prerequisites
- mediumHybrid Search
Re-ranking is typically applied after an initial retrieval step (dense or hybrid) — it's a refinement layer
- hardNLP
Cross-encoders compute pairwise similarity between query and document — understanding embeddings and attention is essential
Recommended reference
Nogueira & Cho (2019) 'Passage Re-ranking with BERT' — arXiv:1901.04085; introduced BERT for re-ranking; pair with Pradeep et al. (2023) for modern approaches
Notes from AI deep research
Anthropic Opus
Cross-encoders na top-k. Nogueira & Cho (2019). Podnosi quality za cene latency
OpenAI Deep Research
Cross-encodery / LLM-as-reranker [OA#34]
Google Deep Think
Wtórne modele sortujące [G#45]
Related skills
- → is part of: Retrieval-Augmented Generation(3/3)
- → is part of: Semantic Search(3/3)