When Hard Negatives Hurt: Bridging the Generative-Discriminative Gap in Hard Negative Synthesis for Retrieval

• URL: http://arxiv.org/abs/2606.01304v2
• Date: Sun, 07 Jun 2026 15:44:02 GMT
• Title: When Hard Negatives Hurt: Bridging the Generative-Discriminative Gap in Hard Negative Synthesis for Retrieval
• Abstract summary: We show that naively incorporating generated negatives into contrastive learning often degrades retrieval performance.<n>Our analysis reveals two compounding failure modes: discriminative-agnostic generation and source-dependent shortcuts.<n>To close this gap, we propose CausalNeg consisting of two main modules.
• Score: 45.5843471557695
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Hard negative mining has become the dominant strategy for training retrievers, yet it faces intrinsic limitations: negatives are bounded by corpus availability, selected by retriever score rather than diagnostic value, and increasingly contaminated by false positives as the retriever improves. LLM-based synthesis offers a principled alternative, where negatives that are unconstrained, targeted, and free from false positive risk. But we show that naively incorporating generated negatives into contrastive learning often degrades retrieval performance. We identify and formalize the root cause as a generative-discriminative gap: LLM generation optimizes for fluent, plausible text, while contrastive learning demands strategic violations of relevance at the decision boundary. Our analysis reveals two compounding failure modes: discriminative-agnostic generation, where the LLM lacks an explicit model of query information needs and defaults to generic or topic-drifted text that provides no contrastive signal; and source-dependent shortcuts, where distributional artifacts enable the model to distinguish negatives by origin rather than relevance, causing gradient drift that actively corrupts optimization. To close this gap, we propose CausalNeg consisting of two main modules: (1) CoT-guided counterfactual perturbation for data construction: decomposes why a document satisfies a query into explicit information requirements, then surgically violates individual requirements to construct negatives with controlled, interpretable hardness. (2) Query-view entropy maximization during training: disperses generated negatives across the similarity spectrum, minimizing the mutual information between source identity and similarity scores to suppress shortcut exploitation. We make our code publicly available at https://github.com/mzhangzhicheng/CausalNeg.

Related papers

• ResRL: Boosting LLM Reasoning via Negative Sample Projection Residual Reinforcement Learning [60.411097231348975]
  Reinforcement Learning with Verifiable Rewards (RLVR) enhances reasoning of Large Language Models (LLMs)<n>This paper proposes negative sample projection Residual Reinforcement Learning (ResRL) that decouples similar semantic distributions among positive and negative responses.
  arXiv  Detail & Related papers  (2026-05-01T03:57:44Z)
• ARHN: Answer-Centric Relabeling of Hard Negatives with Open-Source LLMs for Dense Retrieval [12.84859278829763]
  We propose ARHN (Answer-centric Relabeling of Hard Negatives) to refine hard negative samples using answer-centric relevance signals.<n>We evaluate ARHN on the BEIR benchmark under three configurations: relabeling only, filtering only, and their combination.
  arXiv  Detail & Related papers  (2026-04-13T07:11:30Z)
• Via Negativa for AI Alignment: Why Negative Constraints Are Structurally Superior to Positive Preferences [0.786119752211706]
  Training large language models with negative-only feedback can match or exceed standard reinforcement learning from human feedback.<n>No unified theoretical account explains why negative signals are so effective.<n>Positive preferences and negative constraints are structurally asymmetric.
  arXiv  Detail & Related papers  (2026-03-17T11:52:18Z)
• ReCALL: Recalibrating Capability Degradation for MLLM-based Composed Image Retrieval [64.14282916266998]
  Composed Image Retrieval aims to retrieve target images based on a hybrid query comprising a reference image and a modification text.<n>We propose ReCALL, a model-agnostic framework that follows a diagnose-generate-refine pipeline.<n>Experiments on CIRR and FashionIQ show that ReCALL consistently recalibrates degraded capabilities and achieves state-of-the-art performance.
  arXiv  Detail & Related papers  (2026-02-02T04:52:54Z)
• UniME-V2: MLLM-as-a-Judge for Universal Multimodal Embedding Learning [101.62386137855704]
  We present a novel Universal Multimodal Embedding (UniME-V2) model.<n>Our approach first constructs a potential hard negative set through global retrieval.<n>We then introduce the MLLM-as-a-Judge mechanism, which utilizes MLLMs to assess the semantic alignment of query-candidate pairs.<n>These scores serve as a foundation for hard negative mining, mitigating the impact of false negatives and enabling the identification of diverse, high-quality hard negatives.
  arXiv  Detail & Related papers  (2025-10-15T13:07:00Z)
• RRRA: Resampling and Reranking through a Retriever Adapter [0.0]
  We propose a learnable adapter module that monitors Bi-Encoder representations to estimate the likelihood that a hard negative is actually a false negative.<n>This probability is modeled dynamically and contextually, enabling fine-grained, query specific judgments.<n> Empirical results on standard benchmarks show that our adapter-enhanced framework consistently outperforms strong Bi-Encoder baselines.
  arXiv  Detail & Related papers  (2025-08-07T08:59:57Z)
• Negating Negatives: Alignment with Human Negative Samples via Distributional Dispreference Optimization [37.8788435790632]
  Large language models (LLMs) have revolutionized the role of AI, yet pose potential social risks. Existing methods rely on high-quality positive-negative training pairs, suffering from noisy positive responses that are barely distinguishable from negative ones. We propose Distributional Dispreference Optimization (D$2$O), which maximizes the discrepancy between dispreferred responses and the generated non-negative ones.
  arXiv  Detail & Related papers  (2024-03-06T03:02:38Z)
• AMRFact: Enhancing Summarization Factuality Evaluation with AMR-Driven Negative Samples Generation [57.8363998797433]
  We propose AMRFact, a framework that generates perturbed summaries using Abstract Meaning Representations (AMRs) Our approach parses factually consistent summaries into AMR graphs and injects controlled factual inconsistencies to create negative examples, allowing for coherent factually inconsistent summaries to be generated with high error-type coverage.
  arXiv  Detail & Related papers  (2023-11-16T02:56:29Z)
• Making Large Language Models Better Reasoners with Alignment [57.82176656663245]
  Reasoning is a cognitive process of using evidence to reach a sound conclusion. Recent studies reveal that fine-tuning LLMs on data with the chain of thought (COT) reasoning process can significantly enhance their reasoning capabilities. We introduce an textitAlignment Fine-Tuning (AFT) paradigm, which involves three steps.
  arXiv  Detail & Related papers  (2023-09-05T11:32:48Z)
• Relation-aware Graph Attention Model With Adaptive Self-adversarial Training [29.240686573485718]
  This paper describes an end-to-end solution for the relationship prediction task in heterogeneous, multi-relational graphs. We particularly address two building blocks in the pipeline, namely heterogeneous graph representation learning and negative sampling. We introduce a parameter-free negative sampling technique -- adaptive self-adversarial (ASA) negative sampling.
  arXiv  Detail & Related papers  (2021-02-14T16:11:56Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.