SLIDERS: Structured Reasoning for Scalable QA over Long Documents
A team of researchers has introduced SLIDERS, a novel framework designed for answering questions across extensive document collections. This system extracts key information into a relational database, facilitating scalable reasoning through SQL rather than relying on concatenated text. To maintain overall coherence, a data reconciliation phase employs provenance and extraction ratios. This innovative method effectively tackles the aggregation limitations found in chunk-based approaches.
Key facts
- arXiv:2604.22294v1
- SLIDERS framework for question answering
- Extracts information into relational database
- Uses SQL for reasoning
- Includes data reconciliation stage
- Addresses aggregation bottleneck
- Designed for long document collections
- Provenance and extraction ration used
Entities
—