Autonomous Knowledge Graph Exploration with Adaptive Breadth-Depth Retrieval

ACL ARR 2026 January Submission5387 Authors

Keywords: Knowledge graph retrieval, agentic retrieval, adaptive breadth–depth trade-off, multi-hop graph traversal, tool-using language models, global lexical search, neighborhood exploration, training-free retrieval, heterogeneous knowledge graphs, STaRK benchmark, retrieval efficiency, trajectory distillation, label-free imitation learning, parallel agents

Abstract: Retrieving evidence for language model queries from knowledge graphs requires balancing broad search across the graph with targeted multi-hop traversal to follow relational links. Similarity-based retrievers provide coverage but remain shallow, whereas traversal-based methods rely on selecting seed nodes to start exploration, which can fail when queries span multiple entities and relations. We introduce DGR: Dynamic Graph Retriever, an agentic KG retriever that gives a language model control over this breadth-depth tradeoff using a two-operation toolset: global lexical search over node descriptors and one-hop neighborhood exploration that composes into multi-hop traversal. DGR alternates between breadth-oriented discovery and depth-oriented expansion without depending on a fragile seed selection, a pre-set hop depth, or graph training. DGR autonomously adapts tool use to queries, using global search for language-heavy queries and neighborhood exploration for relation-heavy queries. On STaRK, DGR reaches 59.1% average Hit@1, improving Hit@1 by up to 9.5% and MRR by up to 7.5% over trained retrievers and agentic baselines while maintaining performance on various graph regimes. Finally, we distill DGR tool-use trajectories from a large teacher into an 8B model via label-free imitation, improving Hit@1 by 7.0%, 26.6%, and 13.5% over the base 8B model on AMAZON, MAG, and PRIME datasets, and retaining up to 98.5% of the teacher's Hit@1 rate.

Paper Type: Long

Research Area: Information Extraction and Retrieval

Research Area Keywords: knowledge graphs, graphs, semi-structured databases, knowledge bases, passage retrieval, dense retrieval, document representation, re-ranking, knowledge base construction, evaluation and metrics

Contribution Types: NLP engineering experiment, Publicly available software and/or pre-trained models, Data analysis

Languages Studied: English

Submission Number: 5387