Hugging Face Daily Papers · May 29, 2026 · 5 min read

Is Position Bias in Dense Retrievers Built In-or Learned from Data?

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Like Read original ↗

This is an automated message from the <a href=\"https://huggingface.co/librarian-bots\">Librarian Bot</a>. I found the following papers similar to this paper. \nThe following papers were recommended by the Semantic Scholar API \n<ul>\n<li><a href=\"https://huggingface.co/papers/2605.24236\">MeVer at CheckThat! 2026: Cluster-Aware Hard-Negative Mining for Multilingual Scientific-Source Retrieval</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2604.19047\">RARE: Redundancy-Aware Retrieval Evaluation Framework for High-Similarity Corpora</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2604.11092\">ARHN: Answer-Centric Relabeling of Hard Negatives with Open-Source LLMs for Dense Retrieval</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2604.06097\">Masking or Mitigating? Deconstructing the Impact of Query Rewriting on Retriever Biases in RAG</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2604.03642\">LLM-based Listwise Reranking under the Effect of Positional Bias</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2605.27105\">Lost in the Evidence? Reproducing Document Position and Context Size Effects in RAG</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2605.20683\">Layer-wise Token Compression for Efficient Document Reranking</a> (2026)</li>\n</ul>\n Please give a thumbs up to this comment if you found it helpful!\n If you want recommendations for any Paper on Hugging Face checkout <a href=\"https://huggingface.co/spaces/librarian-bots/recommend_similar_papers\">this</a> Space\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: <code><a href=\"/librarian-bot\">@librarian-bot</a> recommend</code>\n","updatedAt":"2026-05-30T01:43:05.323Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":359,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7190007567405701},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.26578","authors":[{"_id":"6a191dbc56b4bb14ec65d075","name":"Daegon Yu","hidden":false},{"_id":"6a191dbc56b4bb14ec65d076","user":{"_id":"665868596e4d9a66a1b2c779","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/ZiZpk_ZEjqSoBEj6rWgI_.png","isPro":false,"fullname":"SeungYoon Han","user":"seungyoonee","type":"user","name":"seungyoonee"},"name":"SeungYoon Han","status":"claimed_verified","statusLastChangedAt":"2026-05-29T08:49:53.785Z","hidden":false},{"_id":"6a191dbc56b4bb14ec65d077","name":"Woomyoung Park","hidden":false}],"publishedAt":"2026-05-26T00:00:00.000Z","submittedOnDailyAt":"2026-05-29T00:00:00.000Z","title":"Is Position Bias in Dense Retrievers Built In-or Learned from Data?","submittedOnDailyBy":{"_id":"665868596e4d9a66a1b2c779","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/ZiZpk_ZEjqSoBEj6rWgI_.png","isPro":false,"fullname":"SeungYoon Han","user":"seungyoonee","type":"user","name":"seungyoonee"},"summary":"Dense retrievers exhibit positional bias, favoring documents whose query-relevant information appears near the beginning and degrading retrieval performance when the information appears later. While prior work on positional bias in dense retrievers has largely focused on architectural explanations, we study how the positional distribution of evidence in training data affects retrieval-level bias direction. To test this, we construct synthetic position-targeted training sets in which query-relevant evidence appears at the beginning, middle, or end of documents, and fine-tune eight architecturally diverse pretrained models under position-skewed and balanced training distributions. At the ranking level, we observe a strong directional pattern across the examined models: skewed training distributions favor evidence at the corresponding positions. Position-balanced training reduces positional sensitivity by 57--87\\% on position-aware benchmarks, with competitive mean retrieval performance in our controlled setting. Representation-level analyses further suggest that fine-tuning often reshapes learned positional preferences, although pre-existing architectural or pretraining-specific tendencies persist in some models. These results identify training-position distribution as a major controllable factor in retrieval-level position bias and suggest balanced data curation as a practical mitigation strategy.","upvotes":11,"discussionId":"6a191dbc56b4bb14ec65d078","ai_summary":"Training data position distribution significantly influences positional bias in dense retrievers, with balanced training reducing sensitivity by up to 87% while maintaining competitive retrieval performance.","ai_keywords":["dense retrievers","positional bias","query-relevant information","training data distribution","fine-tuning","pretrained models","position-skewed training","position-balanced training","retrieval performance","representation-level analysis"],"organization":{"_id":"64ef692011e7dfe955883ed1","name":"sionic-ai","fullname":"sionic-ai","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/64ef68e04a9ce403b210f307/s9Rcsvnjz1v2HunQy9rY4.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64664217e9906a259f33dd92","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64664217e9906a259f33dd92/rnTr9-OnmKOLk6Mdc8el8.png","isPro":false,"fullname":"kangsungwoo","user":"KSWKSM","type":"user"},{"_id":"633b1f6a0d68f86e2d9528cd","avatarUrl":"/avatars/ed4029192b8e101b1e2c9dd7a9f980dc.svg","isPro":false,"fullname":"정세민","user":"devtorry","type":"user"},{"_id":"68ee45cb9a6f942b016a3dcd","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/68ee45cb9a6f942b016a3dcd/Pwg429hBM6Mg24Cbneie0.jpeg","isPro":false,"fullname":"Joon","user":"mconcat","type":"user"},{"_id":"642b0c2fecec03b4464a1d9b","avatarUrl":"/avatars/76c794c7c17503ae9a0d5336395bd132.svg","isPro":false,"fullname":"KuKu","user":"dragonkue","type":"user"},{"_id":"665868596e4d9a66a1b2c779","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/ZiZpk_ZEjqSoBEj6rWgI_.png","isPro":false,"fullname":"SeungYoon Han","user":"seungyoonee","type":"user"},{"_id":"64ec4c04c782d648d28d70fc","avatarUrl":"/avatars/6975526fcf4b513cc934b5bc45370a48.svg","isPro":false,"fullname":"Sukmin Cho","user":"zomss","type":"user"},{"_id":"6456e8934a7ffb7d5a4b6b45","avatarUrl":"/avatars/f022da2b8d5036a5ad0c19001220de97.svg","isPro":false,"fullname":"de","user":"kaki-paper","type":"user"},{"_id":"67c381e5de068b30716bcf14","avatarUrl":"/avatars/bc562fb76d34d65dc1269786226b30fe.svg","isPro":false,"fullname":"Jeong","user":"Kindred1000","type":"user"},{"_id":"6a194400c08aa4869ef28934","avatarUrl":"/avatars/1e835c2e60fdc51e4ac80456c2867970.svg","isPro":false,"fullname":"kim","user":"bukyo","type":"user"},{"_id":"6a19445a0ccf9357520d23ab","avatarUrl":"/avatars/6a18ae7d47508ff415754c6299e70093.svg","isPro":false,"fullname":"shimsewon","user":"shim111111","type":"user"},{"_id":"65a4c4ed2548c41ad9b1421c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65a4c4ed2548c41ad9b1421c/bMQbowjHKvq-bKpzalvWm.jpeg","isPro":false,"fullname":"Youngjoon Jang","user":"yjoonjang","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"64ef692011e7dfe955883ed1","name":"sionic-ai","fullname":"sionic-ai","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/64ef68e04a9ce403b210f307/s9Rcsvnjz1v2HunQy9rY4.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.26578.md"}">

Papers

arxiv:2605.26578

Is Position Bias in Dense Retrievers Built In-or Learned from Data?

Published on May 26

· Submitted by

SeungYoon Han on May 29

sionic-ai

Upvote

Authors:

SeungYoon Han ,

Abstract

Training data position distribution significantly influences positional bias in dense retrievers, with balanced training reducing sensitivity by up to 87% while maintaining competitive retrieval performance.

AI-generated summary

Dense retrievers exhibit positional bias, favoring documents whose query-relevant information appears near the beginning and degrading retrieval performance when the information appears later. While prior work on positional bias in dense retrievers has largely focused on architectural explanations, we study how the positional distribution of evidence in training data affects retrieval-level bias direction. To test this, we construct synthetic position-targeted training sets in which query-relevant evidence appears at the beginning, middle, or end of documents, and fine-tune eight architecturally diverse pretrained models under position-skewed and balanced training distributions. At the ranking level, we observe a strong directional pattern across the examined models: skewed training distributions favor evidence at the corresponding positions. Position-balanced training reduces positional sensitivity by 57--87\% on position-aware benchmarks, with competitive mean retrieval performance in our controlled setting. Representation-level analyses further suggest that fine-tuning often reshapes learned positional preferences, although pre-existing architectural or pretraining-specific tendencies persist in some models. These results identify training-position distribution as a major controllable factor in retrieval-level position bias and suggest balanced data curation as a practical mitigation strategy.