Bit-factorized DocIDs boost unseen-document generative retrieval by 20 pts without retraining
_reachsumit · x · 2026-10-14
A paper from Leiden University proposes bit-factorized numeric document identifiers for generative retrieval, where each bit's meaning derives from the language model's own representations rather than memorized indexing-stage embeddings.
- Unseen documents can be scored without retraining, fixing the poor generalization of numeric DocIDs
- Over 20 percentage points higher Hit@10 on unseen documents across three benchmarks
- Up to 7x lower inference latency than existing numeric DocID designs on NQ320K and TriviaQA
- Only 3 percentage points below weight-update methods on MS MARCO
More from Research
- AIPOCH's Open-Science ships portable .science packages that share full research context, not just outputs — ahuja_priyank · 2026-10-14
- Video-Arena: Game-Based Video Benchmark Shows Native Models Score Just 13.2 F1 — Kuan Zhang · 2026-10-14
- LSCBench: Vibe Coding Agents Write Functional Code, But Only 25% Is Also Secure — Qwen · 2026-10-14
- Omni-LLMs Know More Than They Say: NAVER's OmniCatalyst Adds 7.29% via Self-Generated Text — naver-ai · 2026-10-14
- Repeated-Data Pretraining Study: Smaller Batches Act as Regularizers and Beat More GPUs — Sewon Min · 2026-10-14
- NAVER's WATCH: Cross-Video Human-Video Pretraining Lifts Robot Success Rate to 77.7% — naver-ai · 2026-10-14