DualGraph: A Hybrid RAG Framework for Semi-Structured QA
A new study introduces DualGraph, a framework designed for Retrieval-Augmented Generation (RAG) aimed at semi-structured question answering. This framework visualizes documents through two connected views: a Textual Knowledge Graph for semantic searches and a Symbolic Knowledge Graph for querying subject-predicate-object triples. It provides different strategies for selecting or combining both types of evidence. Furthermore, the authors created SpecsQA, a benchmark derived from a commercial shopping site, to evaluate performance on semi-structured datasets. You can find the research on arXiv under the identifier 2605.27164.
Key facts
- DualGraph is a RAG framework for semi-structured question answering.
- It uses a Textual Knowledge Graph for semantic retrieval.
- It uses a Symbolic Knowledge Graph for symbolic querying.
- The framework combines semantic and symbolic evidence.
- SpecsQA is a new benchmark from a commercial shopping website.
- The paper is on arXiv with ID 2605.27164.
- The approach addresses limitations of semantic-only retrieval on semi-structured data.
- Symbolic approaches support exact filtering and aggregation but are brittle on noisy text.
Entities
Institutions
- arXiv