Reproduce “Less is More: Data-Efficient Complex Question Answering over Knowledge Bases” 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
Yuncheng Hua, Yuan-Fang Li, Guilin Qi, Wei Wu, Jingyao Zhang, Daiqing Qi
Researchers developed NS-CQA, a neural-symbolic framework, to answer complex questions over knowledge bases with high data efficiency. The model achieved state-of-the-art macro F1 scores of 71.01% on the CQA dataset and 72.04% on WebQuestionsSP, utilizing approximately 1% of the training data compared to prior methods.
Key insights
- Data efficiency for complex KBQA can be substantially improved by providing dense, adaptive feedback and strategically balancing exploration and exploitation in reinforcement learning.
- Reducing the effective vocabulary and search space through masking and copy mechanisms is critical for accelerating convergence and improving generalizability with limited data.
- A carefully designed, simple set of primitive actions combined with basic memory (key-value dictionary) is sufficient for effective complex reasoning over knowledge bases.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.