Dual-View Training for Instruction-Following Information Retrieval

Hugging Face Daily Papers Papers

Summary

A dual-view data synthesis method using polarity reversal boosts instruction-following retrieval performance by 45% on the FollowIR benchmark.

Instruction-following information retrieval (IF-IR) studies retrieval systems that must not only find documents relevant to a query, but also obey explicit user constraints such as required attributes, exclusions, or output preferences. However, most retrievers are trained primarily for semantic relevance and often fail to distinguish documents that match the topic from those that satisfy the instruction. We propose a dual-view data synthesis strategy based on polarity reversal: given a query, a document that is relevant under the instruction, and a hard negative that matches the query but violates the instruction, we prompt an LLM to generate a complementary instruction under which the two documents swap relevance labels. By presenting the same document pair under complementary instructions that invert their relevance labels, the training signal forces the retriever to reconsider the same candidate set through the instruction, rather than relying on fixed topical cues. On a 305M-parameter encoder, our method improves performance on the FollowIR benchmark by 45%, surpassing general-purpose embedding models of comparable or larger scale. Through head-to-head comparisons at matched data budgets, we further show that data diversity and instruction supervision play complementary roles: the former preserves general retrieval quality, while the latter improves instruction sensitivity. These results highlight the value of targeted data synthesis for building retrieval systems that are both broadly capable and instruction-aware.
Original Article
View Cached Full Text

Cached at: 04/22/26, 06:17 AM

Paper page - Dual-View Training for Instruction-Following Information Retrieval

Source: https://huggingface.co/papers/2604.18845

Abstract

A dual-view data synthesis approach using polarity reversal enhances retrieval systems’ ability to follow instructions by training models to distinguish between topic-relevant and instruction-compliant documents.

Instruction-following information retrieval(IF-IR) studiesretrieval systemsthat must not only find documents relevant to a query, but also obeyexplicit user constraintssuch asrequired attributes,exclusions, oroutput preferences. However, most retrievers are trained primarily for semantic relevance and often fail to distinguish documents that match the topic from those that satisfy the instruction. We propose a dual-viewdata synthesisstrategy based onpolarity reversal: given a query, a document that is relevant under the instruction, and a hard negative that matches the query but violates the instruction, we prompt anLLMto generate acomplementary instructionunder which the two documents swaprelevance labels. By presenting the same document pair undercomplementary instructions that invert theirrelevance labels, thetraining signalforces the retriever to reconsider the same candidate set through the instruction, rather than relying on fixed topical cues. On a 305M-parameterencoder, our method improves performance on theFollowIR benchmarkby 45%, surpassinggeneral-purpose embedding modelsof comparable or larger scale. Through head-to-head comparisons at matched data budgets, we further show thatdata diversityandinstruction supervisionplay complementary roles: the former preserves general retrieval quality, while the latter improves instruction sensitivity. These results highlight the value of targeteddata synthesisfor buildingretrieval systemsthat are both broadly capable and instruction-aware.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2604\.18845

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2604.18845 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2604.18845 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2604.18845 in a Space README.md to link it from this page.

Collections including this paper1

Similar Articles

Boosting Visual Instruction Tuning with Self-Supervised Guidance

Hugging Face Daily Papers

This paper proposes augmenting visual instruction tuning in multimodal language models with self-supervised tasks expressed as natural language instructions, improving vision-centric reasoning without additional architecture or annotations. By reformulating classical self-supervised pretext tasks as image-instruction-response triplets, the method achieves consistent performance improvements across multiple benchmarks by injecting only 3-10% visually grounded instructions into the training data.

VCIFBench: Evaluating Complex Instruction Following for Video Understanding

arXiv cs.CL

VCIFBench is a new benchmark for evaluating complex instruction following in video understanding, featuring 306 test instructions with content, format, style, and structure constraints, plus a DPO preference dataset. Experiments on 10 MLLMs reveal that joint constraint satisfaction remains challenging, and DPO training on the benchmark data improves instruction-following performance.