EXCISE: Query-Side Exclusion for Late-Interaction Retrieval

2026-08-05

AI Summary

This work addresses the exclusion reversal problem in late-interaction retrieval models, where the MaxSim mechanism erroneously promotes documents containing excluded terms (e.g., in queries of the form “X but not Z”). The authors propose the first purely query-side, lightweight solution that requires neither modification of frozen indexes nor fine-tuning of the backbone model. Their approach employs two small neural modules (totaling 1.5M parameters) to identify exclusion intents and re-embed the candidate shortlist, combined with a parameter-free rule to downweight matching documents. Evaluated across six datasets and three backbone architectures, the method substantially outperforms both frozen and fine-tuned baselines, boosting ExcluIR’s Success@10 from 0.058 to 0.691 and achieving Boolean NOT accuracy of 0.90–0.92—surpassing even fine-tuned cross-encoders.

Abstract

Late-interaction retrievers handle exclusion queries poorly. When a user asks for X but not Z, the additive MaxSim score promotes documents covering Z, a problem we call exclusion inversion. We show that no readout of the frozen vectors recovers the constraint, because the difficulty lies in identifying the excluded topic, which depends on the query alone. EXCISE operates at query time and corrects the inversion while leaving the index frozen. Two query-side modules totalling 1.5M parameters identify the topic and re-embed a 100-document shortlist, and a parameter-free rule demotes candidates matching that topic. Across six collections and three backbones, EXCISE is the strongest system in all eighteen backbone-collection cells against that backbone's own frozen and fine-tuned baselines. It raises exclusion success@10 on ExcluIR from 0.058 to 0.691 and raises Boolean NOT accuracy from 0.25-0.29 to 0.90-0.92. Pooled over 1,860 queries, it outperforms every fine-tuned cross-encoder, each of which loses no-harm nDCG@10, whereas EXCISE matches its frozen baseline on its strongest backbone. We release X-BENCH, a tiered benchmark of explicit, implicit, and compound exclusions with no-harm and Boolean controls.

Problem

• late-interaction retrieval
• exclusion queries
• exclusion inversion
• query-side exclusion
• retrieval

Innovation

• late-interaction retrieval
• exclusion queries
• query-side correction
• frozen index
• EXCISE