mlsanityData to 14 Oct 2026

Beyond Single-Negative Preference: Multi-Negative DPO for LLM-Centric Historical Entity Linking

arXiv 2609.07379 ↗First posted 7 Sep 2026cs.CL cs.AIcited by 0

On the map: LLM alignment / RLHF / preference optimisation → LLM preference optimization methods

Abstract (arXiv)

Large language models (LLMs) have recently shown promise for historical entity linking, but preference optimization for this task is often formulated with only one negative candidate per training instance. This discards information from the remaining candidates retrieved for the same mention. We introduce multi-negative direct preference optimisation (MDPO), a reference-based pairwise objective that compares the correct entity with all valid rejected candidates associated with each mention. MDPO preserves the Bradley-Terry formulation of DPO while exploiting the complete candidate set through masked, length-normalised sequence scores. We evaluate MDPO on hipe-2020 and newseye, covering French, German, English, Swedish, and Finnish historical newspaper text. Experiments show that MDPO improves over supervised fine-tuning and single-negative DPO, with particularly strong gains for NIL mentions, semantic ambiguity, OCR noise, and historically difficult names. Further analyses disentangle candidate-generation and selection errors, showing that candidate retrieval remains a key bottleneck for end-to-end entity linking. These results demonstrate that incorporating all within-instance negative candidates is a simple and effective improvement for LLM-based historical entity linking.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in LLM preference optimization methods, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 13 of its 34 references that are in the map. First posted on arXiv 7 Sep 2026.

Builds on

13 of 34 references.

Closest ancestors · 3

Same direction, newest first

  1. 2024MPPO: Multi Pair-wise Preference Optimization for LLMs with Arbitrary Negative Samples13 Dec
  2. 2024SimPO: Simple Preference Optimization with a Reference-Free Reward23 May
  3. 2024LiPO: Listwise Preference Optimization through Learning-to-Rank2 Feb

Other references · 10

  1. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  2. 2020Autoregressive Entity Retrievalcited by 334
  3. 2019Scalable Zero-shot Entity Linking with Dense Entity Retrievalcited by 85
  4. 2020Efficient One-Pass End-to-End Entity Linking for Questionscited by 76
  5. 2022ReFinED: An Efficient Zero-shot-capable Approach to End-to-End Entity Linkingcited by 74
  6. 2024LLMAEL: Large Language Models are Good Context Augmenters for Entity Linkingcited by 19
  7. 2023Instructed Language Models with Retrievers Are Powerful Entity Linkerscited by 18
  8. 2023Multilingual End to End Entity Linkingcited by 8
  9. 2025Leveraging the Power of Large Language Models in Entity Linking via Adaptive Routing and Targeted Reasoningcited by 4
  10. 2023Evaluating end-to-end entity linking on domain-specific knowledge bases: Learning about ancient technologies from museum collectionscited by 3

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Hybrid Retrieval-Augmented Generation for Robust Multilingual Document Question Answering
  2. 2025FiNERweb: Datasets and Artifacts for Scalable Multilingual Named Entity Recognition
  3. 2024Learning from Natural Language Explanations for Generalizable Entity Matching
  4. 2023DAMO-NLP at SemEval-2023 Task 2: A Unified Retrieval-augmented System for Multilingual Named Entity Recognition
  5. 2026HELEA: Hard-Negative Benchmark and LLM-based Reranking for Robust Entity Alignment
  6. 2024Smaug: Fixing Failure Modes of Preference Optimisation with DPO-Positive
  7. 2025On Evaluating the Adversarial Robustness of Foundation Models for Multimodal Entity Linking ↗
  8. 2025In-context Ranking Preference Optimization