MULTI3IR Benchmark: 104.9K Multi-perspective Multi-modal Queries
_reachsumit · x · 2026-09-01
MULTI3IR introduces a benchmark of 104.9K Stack Exchange queries annotated with implicit perspective descriptions to address the lack of multi-domain and multi-modal coverage in existing IR benchmarks. The paper also proposes SPIN, a parameter-efficient method that learns noise vectors to steer embeddings towards diverse semantic directions. Experiments show SPIN substantially improves perspective coverage and generalizes well, revealing a single-perspective bias in current multimodal retrievers.
More from Research
- Researcher claims models do decide to start hacking on their own — voooooogel · 2026-09-01
- Discussion: RL instills model behaviors independent of system prompts — voooooogel · 2026-09-01
- Why "it feels better" isn't good enough for production LLM decisions — camerongreen95 · 2026-09-01
- Abliteration technique removes model refusals while keeping coding/cyber capabilities, sparking debate — aryaman2020 · 2026-09-01
- Explanation of Denoising Diffusion Models and Score Matching — ariG23498 · 2026-09-01
- LightRAG: Simple and Fast Retrieval-Augmented Generation — goyalshaliniuk · 2026-09-01