HELIOS: Harmonizing Early Fusion, Late Fusion, and LLM Reasoning for Multi-Granular Table-Text Retrieval
Sungho Park, Joohyung Yun, Jongwuk Lee, Wook-Shin Han
Abstract
Table-text retrieval aims to retrieve relevant tables and text to support open-domain question answering. Existing studies use either early or late fusion, but face limitations. Early fusion pre-aligns a table row with its associated passages, forming "stars," which often include irrelevant contexts and miss query-dependent relationships. Late fusion retrieves individual nodes, dynamically aligning them, but it risks missing relevant contexts. Both approaches also struggle with advanced reasoning tasks, such as column-wise aggregation and multihop reasoning. To address these issues, we propose HELIOS, which combines the strengths of both approaches. First, the edge-based bipartite subgraph retrieval identifies finer-grained edges between table segments and passages, effectively avoiding the inclusion of irrelevant contexts. Then, the query-relevant node expansion identifies the most promising nodes, dynamically retrieving relevant edges to grow the bipartite subgraph, minimizing the risk of missing important contexts. Lastly, the star-based LLM refinement performs logical inference at the star graph level rather than the bipartite subgraph, supporting advanced reasoning tasks. Experimental results show that HELIOS outperforms state-of-the-art models with a significant improvement up to 42.6% and 39.9% in recall and nDCG, respectively, on the OTT-QA benchmark. Love Me Like You Used To (song) Love Me Like You Used To is a song written by Paul Davis and Bobby Emmons, and recorded by American country music artist Johnny Cash.
Ask about this paper
Your agent reads all of it.
Lune indexed this paper to the last equation, along with the top-tier papers that cite it. Ask a question and the answer quotes them.
Cited by top-tier papers3
- Table Question Answering in the Era of Large Language Models: A Comprehensive Survey of Tasks, Methods, and EvaluationWei Zhou, Bolei Ma, Annemarie Friedrich, Mohsen MesgarACL 2026 · 3 citations
- SPARTA: Scalable and Principled Benchmark of Tree-Structured Multi-hop QA over Text and TablesSungho Park, Jueun Kim, Wook-Shin HanICLR 2026 · 2 citations
- LILaC: Late Interacting in Layered Component Graph for Open-domain Multimodal Multihop RetrievalJoohyung Yun, Doyup Lee, Wook-Shin HanEMNLP 2025
Builds on10
- Self-Consistency Improves Chain of Thought Reasoning in Language ModelsXuezhi Wang, Jason Wei, Dale Schuurmans, Quoc V. Le et al.ICLR 2023 · 681 citations
- Chain-of-Table: Evolving Tables in the Reasoning Chain for Table UnderstandingZilong Wang, Hao Zhang, Chun-Liang Li, Julian Martin Eisenschlos et al.ICLR 2024 · 244 citations
- Answering Complex Open-Domain Questions with Multi-Hop Dense RetrievalWenhan Xiong, Xiang Lorraine Li, Srini Iyer, Jingfei Du et al.ICLR 2021 · 232 citations
- MultiModalQA: complex question answering over text, tables and imagesAlon Talmor, Ori Yoran, Amnon Catav, Dan Lahav et al.ICLR 2021 · 229 citations
- Open Question Answering over Tables and TextWenhu Chen, Ming-Wei Chang, Eva Schlinger, William Yang Wang et al.ICLR 2021 · 76 citations
Related papers
- TableRAG: A Retrieval Augmented Generation Framework for Heterogeneous Document ReasoningXiaohan Yu, Pu Jian, Chong ChenEMNLP 2025 · 4 citations
- Is Table Retrieval a Solved Problem? Exploring Join-Aware Multi-Table RetrievalPeter Baile Chen, Yi Zhang, Dan RothACL 2024 · 4 citations
- REaR : Retrieve, Expand and Refine for Effective Multitable RetrievalRishita Agarwal, Himanshu Singhal, Peter Baile Chen, Manan Roy Choudhury et al.ACL 2026 · 2 citations
- Dual Reader-Parser on Hybrid Textual and Tabular Evidence for Open Domain Question AnsweringAlexander Hanbo Li, Patrick Ng, Peng Xu, Henghui Zhu et al.ACL 2021
- RINK: Reader-Inherited Evidence Reranker for Table-and-Text Open Domain Question AnsweringEunhwan Park, Sung-Min Lee, Daeryong Seo, Seonhoon Kim et al.AAAI 2023 · 4 citations
