Learning from Emptiness: De-biasing Listwise Rerankers with Content-Agnostic Probability Calibration
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911585251885056 |
|---|---|
| author | Lv, Hang Gu, Hongchao Yang, Ruiqing Li, Liangyue Chen, Zulong Lian, Defu Wang, Hao Chen, Enhong |
| author_facet | Lv, Hang Gu, Hongchao Yang, Ruiqing Li, Liangyue Chen, Zulong Lian, Defu Wang, Hao Chen, Enhong |
| contents | Generative listwise reranking leverages global context for superior retrieval but is plagued by intrinsic position bias, where models exhibit structural sensitivity to input order independent of relevance. Existing mitigations present a dilemma: inference-time aggregation incurs prohibitive latency, while training-based methods often fail to eradicate ingrained priors, particularly in compact models. To resolve this dilemma, we propose CapCal (Content-Agnostic Probability Calibration), a training-free framework that mechanically decouples positional bias from ranking decisions. By estimating the bias distribution via content-free placeholders, CapCal rectifies output logits through an entropy-adaptive contrastive mechanism. Evaluations across 10 benchmarks confirm that CapCal achieves superior performance among training-free methods while preserving single-pass efficiency. Notably, it unlocks the latent potential of lightweight models (e.g., 0.6B), delivering absolute NDCG gains exceeding 10 points and outperforming both permutation-based aggregation and data-augmentation baselines. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_10150 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Learning from Emptiness: De-biasing Listwise Rerankers with Content-Agnostic Probability Calibration Lv, Hang Gu, Hongchao Yang, Ruiqing Li, Liangyue Chen, Zulong Lian, Defu Wang, Hao Chen, Enhong Artificial Intelligence Computation and Language Generative listwise reranking leverages global context for superior retrieval but is plagued by intrinsic position bias, where models exhibit structural sensitivity to input order independent of relevance. Existing mitigations present a dilemma: inference-time aggregation incurs prohibitive latency, while training-based methods often fail to eradicate ingrained priors, particularly in compact models. To resolve this dilemma, we propose CapCal (Content-Agnostic Probability Calibration), a training-free framework that mechanically decouples positional bias from ranking decisions. By estimating the bias distribution via content-free placeholders, CapCal rectifies output logits through an entropy-adaptive contrastive mechanism. Evaluations across 10 benchmarks confirm that CapCal achieves superior performance among training-free methods while preserving single-pass efficiency. Notably, it unlocks the latent potential of lightweight models (e.g., 0.6B), delivering absolute NDCG gains exceeding 10 points and outperforming both permutation-based aggregation and data-augmentation baselines. |
| title | Learning from Emptiness: De-biasing Listwise Rerankers with Content-Agnostic Probability Calibration |
| topic | Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2604.10150 |