Paper Detail
Antoine Edy, Max Conti, Quentin Macé
While Late Interaction models exhibit strong retrieval performance, many of their underlying dynamics remain understudied, potentially hiding performance bottlenecks. In this work, we focus on two topics in Late Interaction retrieval: a length bias that arises when using multi-vector scoring, and the similarity distribution beyond the best scores pooled by the MaxSim operator. We analyze these behaviors for state-of-the-art models on the NanoBEIR benchmark. Results show that while the theoretical length bias of causal Late Interaction models holds in practice, bi-directional models can also suffer from it in extreme cases. We also note that no significant similarity trend lies beyond the top-1 document token, validating that the MaxSim operator efficiently exploits the token-level similarity scores.
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@misc{edy2026working,
title = {Working Notes on Late Interaction Dynamics: Analyzing Targeted Behaviors of Late Interaction Models},
author = {Antoine Edy and Max Conti and Quentin Macé},
year = {2026},
abstract = {While Late Interaction models exhibit strong retrieval performance, many of their underlying dynamics remain understudied, potentially hiding performance bottlenecks. In this work, we focus on two topics in Late Interaction retrieval: a length bias that arises when using multi-vector scoring, and the similarity distribution beyond the best scores pooled by the MaxSim operator. We analyze these behaviors for state-of-the-art models on the NanoBEIR benchmark. Results show that while the theoretica},
url = {https://huggingface.co/papers/2603.26259},
keywords = {Late Interaction models, multi-vector scoring, MaxSim operator, NanoBEIR benchmark, length bias, token-level similarity scores, huggingface daily},
eprint = {2603.26259},
archiveprefix = {arXiv},
}
{}