Searcharxiv⌕ Search

arXiv · 2611.15417

When Direction Matters: Diagnosing Relational Instruction Following in Dense Retrieval

Abstract

Dense retrievers increasingly claim the ability to condition retrieval on natural-language relevance criteria beyond a short query. We identify a specific weakness in this capability: the evaluated models perform well when lexical or attribute cues distinguish the target, but become much less reliable when relevance depends on a directional relation over the same entities. We introduce SwitchIR, a controlled diagnostic that contrasts Constraint Shift and Explicit Filtering with Relational Flip, where the query and document pair remain fixed while the requested relation direction reverses. Across representative embedding and late-interaction retrievers, the primary joint metric is high on the two cue-available controls and consistently low on Relational Flip; the same ordering persists in an LLM-assisted, manually reviewed realistic variant. An exact margin decomposition shows that both directions are correct only when the direction-sensitive query--document interaction exceeds the direction-invariant pair preference. Empirically, direction-sensitive changes are common, but their aligned interaction is usually too weak to overcome that preference. Diagnostic fine-tuning shows that models can learn relational behavior that generalizes across relation families, with stronger generalization in the larger model.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Zhanyu Wu, Richong Zhang. 2026-11-11. When Direction Matters: Diagnosing Relational Instruction Following in Dense Retrieval. https://searcharxiv.com/abs/2611.15417

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite Searcharxiv by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

Beyond Retrieval: Learning Compact User Representations for Scalable LLM Personalization

Personalizing large language models requires adapting model behavior to individual users while preserving robustness and deployment-scale efficiency. Existing approaches typically personalize LLMs either at the input level, by retrieving user histories or constructing profile prompts, or at the parameter level, by maintaining user-specific parameter-efficient modules. The former makes personalization sensitive to retrieval quality and prompt design, whereas the latter incurs storage and maintenance costs that grow with the user population. To address these limitations, we propose TAP-PER (Temporal Attentive Prefix for PERsonalization), a prefix-based framework that encodes user preferences as learnable representations, avoiding the serialization of user histories into prompts and replacing heavy per-user adapters with lightweight user-state prefix embeddings. Inspired by personalized recommendation systems, TAP-PER decomposes user modeling into user-state and query-conditioned components, and incorporates temporal signals to capture the evolving nature of user interests. Experiments on six LaMP tasks show that TAP-PER consistently outperforms prompt-based and model-based baselines across classification, rating, and generation settings. Moreover, TAP-PER uses 130x fewer per-user parameters than OPPU and roughly half the total parameter footprint of PER-PCS at the 1,000-user scale, demonstrating scalable personalization without prompt-serialized histories or heavy per-user adapters.

cs.IR↗

RecHarness: A Bandit-Routed Agentic Harness for Self-Evolving Recommender Systems

Optimizing recommender models remains heavily reliant on manual iteration over architectures, objectives, and training strategies. While LLM-based agents can automate this trial-and-error process, relying on LLMs to both select modification directions and formulate concrete hypotheses can lead to unstable search under limited trial budgets. Inspired by the above challenge, we propose RecHarness, a Bandit-Routed Agentic Harness for automated recommender model optimization. RecHarness decouples the optimization process into two steps: a bandit router selects the next modification direction according to historical validation feedback, while the LLM generates a concrete optimization hypothesis and executable code edit within the selected direction. To sustain long-horizon exploration, RecHarness uses a jump-basin mechanism to activate a structural-jump arm when local edits stagnate. Across multiple recommendation tasks, datasets, and model backbones, RecHarness achieves more consistent performance improvements and uses limited trial budgets more effectively than LLM-directed search. During a 7-day online A/B test on a large-scale short-video advertising platform, the selected candidate improves ADVV by 2.084%, Revenue by 0.534%, and Exposure by 0.559%. Code is available at: https://github.com/6lyc/RecHarness.

cs.IR↗

Retrieving Complementary Evidence for Multi-Hop RAG with Semantic Gain Graphs

Multi-hop retrieval-augmented generation (RAG) requires finding documents whose evidence is useful in combination. Query relevance and graph connectivity help locate related content, but neither directly measures the additional information one document contributes given another. We introduce SG$^2$RAG, a framework that represents this complementarity through semantic gain. Offline, a large language model assesses the relevant and nonredundant information contributed by one chunk given another within neighborhoods proposed by semantic similarity. These ordered assessments are aggregated into symmetric edge weights to construct a reusable Semantic Gain Graph. Online, controlled Personalized PageRank starts from query-relevant seeds and ranks complementary evidence beyond the initial retrieval set. Experiments on MuSiQue, 2Wiki, and HotpotQA with two language model backbones show that SG$^2$RAG achieves the best average EM, F1, and Recall@10 among the evaluated single-retrieval methods. Under identical candidate support and propagation, semantic gain improves average F1 and Recall@10 over similarity weighting by 4.08 and 4.31 points. Analyses of supporting documents and improvements in three iterative retrieval frameworks further support its effectiveness as a reusable retrieval component.

cs.IR↗