BLAIR benchmark tests 11 LLMs as recommendation encoders, releases 571M-review Amazon dataset
burkov · x · 2026-11-03
General text-embedding leaderboards don't predict how well models work inside recommendation systems, where item descriptions are short/noisy and embeddings feed downstream recommenders. BLAIR addresses this gap by benchmarking 11 language models as semantic encoders across sequential recommendation, collaborative filtering, and product search. It also releases Amazon Reviews 2023 (571M+ reviews, 48M items) and adds a complex-query search task for long natural-language requests, using queries generated from Amazon reviews plus real user queries.
More from Research
- PrimaCons benchmark exposes cross-primitive inconsistency in Jev decision-making — Xu Liu · 2026-11-03
- MACR: confidence-guided relay multi-agent framework cuts token overhead 6x — Xu Liu · 2026-11-03
- MIT's Weight-Space Programming Touted as a Path to True Continual Learning — daniel_mac8 · 2026-11-03
- LLM eval researcher LChoshen joins Weizmann Institute to launch CoLab — LChoshen · 2026-11-03
- Editing 0.016% of weights fixes Qwen reasoning loops, MATH-500 jumps from 7.9% to 48.1% — DeliciousGorilla · 2026-11-03
- MIT's Soft-MTP: one-line fix removes 80% noisy gradient, boosts MTP pre-training for free — kastnerkyle · 2026-11-03