Reproduce “Recursive Question Understanding for Complex Question Answering over Heterogeneous Personal Data” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Philipp Christmann, Gerhard Weikum
REQAP enables complex analytical question answering over an individual's diverse personal data by translating natural language queries into executable operator trees that run efficiently on-device for privacy. The framework, including its small-scale supervised fine-tuned variant, substantially outperforms state-of-the-art RAG and code generation baselines, achieving a Hit@1 score of 0.380 and a relaxed Hit@1 of 0.53 on the new PERQA benchmark.
Key insights
- A hybrid approach combining symbolic reasoning (operator trees) with neural text processing (LLMs for retrieval and extraction) effectively addresses complex analytical QA over heterogeneous personal data.
- Recursive decomposition of complex questions via in-context learning, followed by distillation into smaller models, enables accurate query planning while maintaining a low computational footprint suitable for on-device deployment.
- Dedicated operators for fuzzy retrieval and semantic extraction from unstructured text are crucial for integrating diverse personal data types into a unified query framework.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.