Structured knowledge retrieval improves by co-adapting rules and search strategies
Co-adaptive Rule Routing and Constraining for Structured Knowledge Retrieval
Information RetrievalArtificial Intelligence
Summary
Finding relevant information from organized databases is hard when clear instructions about how to search are missing. The authors created a system called CoRE that learns both how to pick a searching rule for a question and how to find the right answers following that rule, without needing examples of the right rules beforehand. By training two parts together—one that picks rules and one that searches—they improve each other over time. Their approach works well for questions in science, medicine, and shopping data.
What this means in practice
- •For biomedical data teams: Improve retrieval of structured biomedical evidence without predefined query-rule pairs for drug and anatomy reasoning tasks.
- •For e-commerce platform engineers: Enhance product knowledge retrieval by learning dynamic query routing rules to rank relevant product information without manual rule labeling.
Authors
Zhisheng Qi, Haoyu Han, Jiaxi Li, Bing He, Huiyuan Chen, Franck Dernoncourt, Ryan A. Rossi, Mahantesh M. Halappanavar, Jin Lu, Yu Zhang, Tianming Liu, Yu Wang
Abstract
Structured knowledge retrieval acquires task-relevant evidence from structured knowledge sources for downstream tasks. Existing methods rely on unconstrained retrieval or predefined query-rule supervision, such as author-paper rules for scientific discovery and anatomy-drug rules for precision medicine. Such supervision is rarely available in practice. We propose Co-Adaptive Rule Routing and Constraining Retrieval (CoRE), which jointly learns which reasoning rule to assign to each query and how to retrieve evidence under that rule, without predefined query-rule pairs. A routing agent selects a rule, and its corresponding retriever traverses the knowledge base and ranks evidence that satisfies the rule's constraints. Because routing and retrieval are interdependent, CoRE alternates between reinforcement learning for the routing policy and contrastive learning for the rule-constrained retriever. The selected rules provide supervision for retriever training, while retrieval performance rewards the routing policy. This closed-loop process enables both components to co-adapt without explicit query-rule supervision. Experiments on STaRK-QA datasets across scientific discovery, biomedical reasoning, and e-commerce knowledge management demonstrate CoRE's effectiveness. Ablations show the benefit of co-adapting the router and retriever. Generation-based routing transfers across datasets and benefits from multi-domain pretraining followed by target-specific fine-tuning. Analysis of rule usage across queries also informs automatic rule refinement and discovery.