Training for Compositional Sensitivity Reduces Dense Retrieval Generalization

Radoslav Ralev; Aditeya Baral; Iliya Zhechev; Jen Agarwal; Srijith Rajamohan

cs.IRMar 16, 2026

Training for Compositional Sensitivity Reduces Dense Retrieval Generalization

Authors: Radoslav Ralev, Aditeya Baral, Iliya Zhechev, Jen Agarwal, Srijith Rajamohan

Organizations: Redis, Bulgaria and Redis, USA

Abstract

Dense retrieval compresses texts into single embeddings ranked by cosine similarity. While efficient for recall, this interface is brittle for identity-level matching: minimal compositional edits (negation, role swaps) flip meaning yet retain high similarity. Motivated by geometric results for unit-sphere cosine spaces (Kang et al., 2025), we test this retrieval-composition tension in text-only retrieval. Across four dual-encoder backbones, adding structure-targeted negatives consistently reduces zero-shot NanoBEIR retrieval (8-9% mean nDCG@10 drop on small backbones; up to 40% on medium ones), while only partially improving pooled-space separation. Treating pooled cosine as a recall interface, we then benchmark verifiers scoring token--token cosine maps. MaxSim (late interaction) excels at reranking but fails to reject structural near-misses, whereas a small Transformer over similarity maps reliably separates near-misses under end-to-end training.

Figures & tables

	nDCG@10			Acc@1
Backbone	Model A	Model B	$\Delta$ (% drop)	Model A	Model B	$\Delta$ (% drop)
MiniLM-L6	0.439 $\pm$ 0.000	0.401 $\pm$ 0.001	-0.038 (-8.7%)	0.393 $\pm$ 0.002	0.346 $\pm$ 0.004	-0.047 (-12.0%)
MiniLM-L12	0.467 $\pm$ 0.001	0.424 $\pm$ 0.005	-0.043 (-9.2%)	0.424 $\pm$ 0.003	0.369 $\pm$ 0.010	-0.055 (-13.0%)
GTE-Small	0.481 $\pm$ 0.002	0.442 $\pm$ 0.006	-0.039 (-8.1%)	0.444 $\pm$ 0.001	0.389 $\pm$ 0.004	-0.055 (-12.4%)
GTE-ModernBERT-base	0.543 $\pm$ 0.001	0.324 $\pm$ 0.018	-0.219 (-40.3%)	0.493 $\pm$ 0.005	0.275 $\pm$ 0.015	-0.218 (-44.2%)

Table 1: Mean NanoBEIR retrieval performance (nDCG@10 and Acc@1). Model A: standard fine-tuning. Model B: + structured negatives.

Figure 1: Cosine-similarity distributions between an anchor sentence and a minimally edited near-miss under pooled embeddings. We compare Model A vs. Model B for three perturbation families (negation, binding/order, spatial) and overlay NQ positives/negatives for reference (10k pairs each). Lower is better for near-miss distributions.

Figure 2: NanoBEIR performance after reranking top-

K

candidates with

F_{k}

under a frozen-encoder (blue) or end-to-end (orange) regime; horizontal lines show encoder-only baselines (Model A and Model B). MaxSim (

F_{1}

) is the strongest frozen reranker; end-to-end

F_{4}

is most competitive.

Figure 3: Synthetic structural near-miss test. Mean scores on hard negatives (near-misses); lower is better. The dotted line is pooled cosine from Model B. Simple reductions of

M

(

F_{0}

–

F_{2}

) and MaxSim (

F_{1}

) score near-misses as highly similar, while topology-aware verifiers (

F_{3}

F_{4}

) substantially reduce near-miss scores; end-to-end

F_{4}

is strongest on spatial flips.

Information Retrieval

Information Retrieval studies how to find, rank, and organize relevant information in large collections of data.

See all 471 papers

Reproducing Complex Set-Compositional Information Retrieval
May 5, 2026Vincent Degenhart, Dewi Timman, Arjen P. de Vries +2Query Complexity Lexical Retrieval
Radboud University · Nijmegen, NL
Latent Terms: Dense Retrievers Contain Trivially Extractable BM25-ready Zipfian Vocabularies
May 28, 2026Benjamin Clavié, Sean Lee, Aamir Shakir +1Learned Sparse Retrieval Information Retrieval
Mixedbread AI · National Institute of Informatics (NII) · University of Tsukuba
CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker Distillation
Sep 3, 2026Tingyu Song, Mingxin Li, Yanzhao Zhang +5Multimodal Embedding Cross-Modal Retrieval
CASIA · Alibaba Group · University of Chinese Academy of Sciences +1

May 5, 2026J/K move · Enter open · S save

Reproducing Complex Set-Compositional Information Retrieval

Vincent Degenhart, Dewi Timman, Arjen P. de Vries, Faegheh Hasibi+1

Radboud University · Nijmegen, NL

Query Complexity Lexical Retrieval Information Retrieval IR Evaluation Dense Retrieval

Open paper AskPDF arXiv

Complex information needs may involve set-compositional queries using conjunction, disjunction, and exclusion, yet it remains unclear whether current retrieval paradigms genuinely satisfy such constraints or exploit `semantic shortcuts'. We conduct a reproducibility study to benchmark major retrieval families and reasoning-targeted methods on QUEST and QUEST+Variants, and introduce LIMIT+, a controlled benchmark where relevance depends on arbitrary attribute predicates and constraint satisfaction, and less on pretrained knowledge. Our findings show that (i) on QUEST, the best neural retrievers achieve an effectiveness that is more than double what can be achieved with BM25 (Recall@100

{>}

0.41 vs.\ 0.20), but reasoning-targeted methods like ReasonIR and Search-R1 do not outperform general-purpose retrievers uniformly; (ii) on LIMIT+, gains fail to transfer, where the strongest QUEST method collapses from Recall@100

{\approx}

0.42 to below 0.02, while classic lexical retrieval gains to

{\sim}

0.96. Lastly, (iii) stratifying by compositional depth reveals a consistent degradation across all methods, where algebraic sparse and lexical methods show more stable performance while dense approaches collapse. We release code and LIMIT+ data generation scripts to support future reproducibility and controlled evaluation.

Abstract

Information Retrieval

Explore similar work

Reproducing Complex Set-Compositional Information Retrieval

Latent Terms: Dense Retrievers Contain Trivially Extractable BM25-ready Zipfian Vocabularies

CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker Distillation