mlsanityData to 14 Oct 2026

Polaris: Learning to Generate Table Descriptions from Retrieval Feedback

arXiv 2608.17171 ↗First posted 17 Aug 2026cs.CL cs.DBcited by 0

On the map: LLM alignment / RLHF / preference optimisation → Preference alignment for recommenders and RAG

Abstract (arXiv)

Many table-centric NLP tasks such as NL2SQL first retrieve relevant tables from large collections using keyword search. Recent work uses LLMs to generate natural-language table descriptions to improve retrieval, but they are typically optimized for fluency rather than retrieval effectiveness. We present Polaris, a system that trains an LLM to generate table descriptions directly from retrieval feedback. Our key insight is that existing table retrieval benchmarks already contain the supervision needed for this task: given query-table relevance judgments, we generate multiple candidate descriptions for each table, rank them by their BM25 retrieval effectiveness, and use the resulting preference pairs to fine-tune the LLM with Direct Preference Optimization (DPO). Polaris further expands abbreviated table and column names before generation to reduce vocabulary mismatch. Extensive experiments show that Polaris outperforms the state-of-the-art AutoDDG solution, often by a significant margin. More broadly, our results demonstrate that retrieval benchmarks can be repurposed as supervision for training LLMs to generate retrieval-oriented metadata.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Preference alignment for recommenders and RAG, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 29 of its 49 references that are in the map. First posted on arXiv 17 Aug 2026.

Builds on

29 of 49 references.

References · 29

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  3. 2020Dense Passage Retrieval for Open-Domain Question Answeringcited by 3,824
  4. 2023QLoRA: Efficient Finetuning of Quantized LLMscited by 3,418
  5. 2019Fine-Tuning Language Models from Human Preferencescited by 2,150
  6. 2023RLAIF vs. RLHF: Scaling Reinforcement Learning from Human Feedback with AI Feedbackcited by 625
  7. 2023Zephyr: Direct Distillation of LM Alignmentcited by 506
  8. 2021SPLADE: Sparse Lexical and Expansion Model for First Stage Rankingcited by 286
  9. 2019Sherlock: A Deep Learning Approach to Semantic Data Type Detectioncited by 74
  10. 2022An Analysis of Fusion Functions for Hybrid Retrievalcited by 51
  11. 2020Valentine: Evaluating Matching Techniques for Dataset Discoverycited by 35
  12. 2022ReasTAP: Injecting Table Reasoning Skills During Pre-training via Synthetic Reasoning Examplescited by 29
17 more
  1. 2022StruBERT: Structure-aware BERT for Table Search and Matchingcited by 24
  2. 2023CHORUS: Foundation Models for Unified Data Discovery and Explorationcited by 24
  3. 2025Pneuma: Leveraging LLMs for Tabular Data Representation and Retrieval in an End-to-End Systemcited by 23
  4. 2020Table Search Using a Deep Contextualized Language Modelcited by 22
  5. 2018Ad Hoc Table Retrieval using Semantic Similaritycited by 21
  6. 2025TARGET: Benchmarking Table Retrieval for Generative Taskscited by 17
  7. 2021Auctus: A Dataset Search Engine for Data Augmentationcited by 16
  8. 2023NameGuess: Column Name Expansion for Tabular Datacited by 13
  9. 2025ExpandR: Teaching Dense Retrievers Beyond Queries with LLM Guidancecited by 10
  10. 2024Synthetic SQL Column Descriptions and Their Impact on Text-to-SQL Performancecited by 6
  11. 2025Automatic database description generation for Text-to-SQLcited by 5
  12. 2025Columbo: Expanding Abbreviated Column Names for Tabular Data Using Large Language Modelscited by 5
  13. 2020Leveraging Schema Labels to Enhance Dataset Searchcited by 4
  14. 2025Leveraging Retrieval Augmented Generative LLMs For Automated Metadata Description Generation to Enhance Data Catalogscited by 3
  15. 2021WTR: A Test Collection for Web Table Retrievalcited by 2
  16. 2026Less Is More? When Dataset Context Hurts LLM-Generated Dataset Descriptionscited by 1
  17. 2026TACO: Task-Aware Column Description Generation Using LLMscited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Improving Table Understanding with LLMs and Entity-Oriented Search
  2. 2026FT-RAG: A Fine-grained Retrieval-Augmented Generation Framework for Complex Table Reasoning
  3. 2024TableRAG: Million-Token Table Understanding with Language Models
  4. 2025Improving Table Retrieval with Question Generation from Partial Tables
  5. 2024Uncovering Limitations of Large Language Models in Information Seeking from Tables
  6. 2025CRAFT: Training-Free Cascaded Retrieval for Tabular QA
  7. 2025RAG over Tables: Hierarchical Memory Index, Multi-Stage Retrieval, and Benchmarking
  8. 2026FGTR: Fine-Grained Multi-Table Retrieval via Hierarchical LLM Reasoning