Academic

HELIOS: Harmonizing Early Fusion, Late Fusion, and LLM Reasoning for Multi-Granular Table-Text Retrieval

arXiv:2603.02248v1 Announce Type: cross 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 multi-hop 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 edg

S
Sungho Park, Joohyung Yun, Jongwuk Lee, Wook-Shin Han · · 1 min read · 21 views
#cs.DB #cs.CL #cs.IR #cs.LG

arXiv:2603.02248v1 Announce Type: cross 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 multi-hop 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.

Executive Summary

The article proposes HELIOS, a novel approach to multi-granular table-text retrieval, which harmonizes early fusion, late fusion, and LLM reasoning to address the limitations of existing methods. HELIOS achieves significant improvements in recall and nDCG on the OTT-QA benchmark, outperforming state-of-the-art models. The approach combines edge-based bipartite subgraph retrieval, query-relevant node expansion, and star-based LLM refinement to effectively retrieve relevant tables and text, supporting advanced reasoning tasks.

Key Points

  • ▸ HELIOS combines early and late fusion approaches to improve table-text retrieval
  • ▸ The approach uses edge-based bipartite subgraph retrieval to avoid irrelevant contexts
  • ▸ Star-based LLM refinement enables advanced reasoning tasks, such as column-wise aggregation and multi-hop reasoning

Merits

Improved Retrieval Accuracy

HELIOS achieves significant improvements in recall and nDCG, outperforming state-of-the-art models

Effective Context Handling

The approach effectively avoids irrelevant contexts and retrieves relevant edges to grow the bipartite subgraph

Demerits

Computational Complexity

The combination of multiple techniques may increase computational complexity, potentially affecting scalability

Expert Commentary

The proposed HELIOS approach demonstrates a significant advancement in multi-granular table-text retrieval, addressing the limitations of existing methods. The combination of edge-based bipartite subgraph retrieval, query-relevant node expansion, and star-based LLM refinement enables effective retrieval of relevant tables and text, supporting advanced reasoning tasks. However, further research is needed to investigate the computational complexity and scalability of the approach. Overall, HELIOS has the potential to improve the performance of various applications, such as search engines and question answering systems, and may influence the design of future information retrieval systems.

Recommendations

  • ✓ Further research should investigate the computational complexity and scalability of the HELIOS approach
  • ✓ The application of HELIOS to various domains and tasks should be explored to demonstrate its generalizability and effectiveness

Sources

Related Articles