Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2604.05766

The LLM Effect on IR Benchmarks: A Meta-Analysis of Effectiveness, Baselines, and Contamination

Moritz Staudinger·Wojciech Kusa·Allan Hanbury
View original source ↗Read source document ↗Export metadata ↓

Abstract

Benchmark collections have long enabled controlled comparison and cumulative progress in Information Retrieval (IR). However, prior meta-analyses have shown that reported effectiveness gains often fail to accumulate, in part due to the use of weak or outdated baselines. While large language models are increasingly used in retrieval pipelines, their impact on established IR benchmarks has not been systematically analyzed. In this study, we analyze 143 publications reporting results on the TREC Robust04 collection and the TREC Deep Learning 2020 (DL20) passage retrieval benchmark to examine longitudinal trends in retrieval effectiveness and baseline strength. We observe what we term an \emph{LLM effect}: recent systems incorporating LLM components achieve 8.8\% higher nDCG@10 on DL20 compared to the best result from TREC 2020 and approximately 20\% higher on Robust04 since 2023. However, adapting a data contamination detection approach to reranking reveals measurable contamination in both benchmarks. While excluding contaminated topics reduces effectiveness, confidence intervals remain wide, making it difficult to determine whether the LLM effect reflects genuine methodological advances or memorization from pretraining data.

Explore related subjects

cs.IR

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Moritz Staudinger, Wojciech Kusa, Allan Hanbury. 2026-04-07. The LLM Effect on IR Benchmarks: A Meta-Analysis of Effectiveness, Baselines, and Contamination. https://arxiv.org/abs/2604.05766

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

RankSteer: Can Pointwise LLM Rankers Be Calibrated at the Representation Level?

Large language models (LLMs) are strong zero-shot pointwise rankers, but lag behind pairwise and listwise methods. Beyond missing comparative signals, we identify a \textit{calibration gap}: ranking-relevant information encoded in hidden states is not fully captured by the scalar output head. We propose RankSteer, a post-hoc activation-steering framework that calibrates ranking via projection-based interventions along multiple directions at inference time: decision, evidence, and, optionally, role. This is achieved without updating model weights or introducing cross-document comparisons. We instantiate RankSteer on two structurally distinct pointwise variants and observe improvements over their respective baselines on most TREC DL and BEIR datasets across three backbones. This suggests that the calibration gap is a general property of pointwise rankers. Our additional geometric analysis shows that steering improves ranking by concentrating each query's document representations along an existing ranking geometry, offering new insight into how LLMs internally represent and calibrate relevance judgments.

cs.IR↗
arXiv2026

IntHQ: Task-Interactive Hierarchical Query on Dual-Stream Representations for Generative Recommendation

Multi-task learning over heterogeneous data is fundamental to modern recommendation, while generative models are emerging as the backbone of next-generation recommenders. However, the integration of multi-task learning into the generative paradigm remains largely unexplored. Existing multi-task recommenders, in both discriminative and generative paradigms, extract task-relevant features from a single task-agnostic representation and wire tasks into a predefined conversion funnel. We show that this scheme is inherently prone to a threefold collapse. Source collapse, where task-specific signals are injected late and diluted in the shared latent space. Relational collapse, where task dependencies are either implicitly absorbed by the backbone or statically fixed by predefined funnels. Hierarchical collapse, where tasks depend on features at different scales and shift across training stages. We propose IntHQ, a multi-task generative recommender with three components, each alleviating one collapse. Dual-Stream Decoupling (DSD) injects task identity into computation stream early and separates the shared context stream from the task-specific stream, alleviating signal dilution. Task-Interactive Modeling (TIM) replaces the predefined funnel with explicit cross-task interaction, letting each task condition on the realized outcomes of its predecessors with learned, input-adaptive strength. Hierarchical Querying (HQ) lets each task gather multi-scale information across different layers at different training stages. In offline evaluations, IntHQ consistently outperforms competitive encoder backbones under four representative task-head configurations. Deployed in production on Amap, serving hundreds of millions of users for travel recommendation, IntHQ yields a 1.60\% relative UVCTR lift.

cs.IR↗
arXiv2026

Verify, Don't Trust: Agentic Model Development for Video Discovery Retrieval at Scale

Large language model (LLM) agents can propose, implement, and evaluate model changes. Autoresearch loops demonstrate this capability through minutes-scale iterations on a self-contained program. Online autoresearch instead spans asynchronous systems, hours-long variants, and weeks-long campaigns that can influence a product. A completed run can still support an invalid conclusion when a code change is a no-op, data windows leak, evaluator semantics drift, or the two arms traverse different serving funnels. We present EvoPilot, a human-gated method for long-horizon online autoresearch. Role-specific agents execute each round through a versioned domain skill and typed adapter. Durable records preserve experiments and failures; deterministic checks enforce recorded lessons. We study a 37-day campaign for the retrieval system that powers Video Deep Dive (VDD), an online experience for discovering follow-on videos after a user opens a seed video. The campaign covered seven directions and used an hourly refreshed index of hundreds of millions of videos. Earlier manual experiments had not established a benefit from an interaction head. A primitive autoresearch attempt revisited the direction but incorrectly attributed an offline hit-rate decline of 22 percentage points to the head. We then introduced EvoPilot. Its human-gated verification traced the drop to a pre-existing evaluation defect that produced output depths of 3,000 and 600. After repair, a matched comparison measured an offline improvement of 3.20 percentage points. Post-study replay and mutation tests rejected invalid comparisons while admitting valid counterparts. Durable state recovered an interrupted round, and artifact reuse avoided approximately five GPU-hours. Separately, a seven-day randomized online evaluation estimated a 0.66% relative increase in the VDD slice of Good Search Result Rate for Retention (GSRR).

cs.IR↗