SciNets: Graph-Constrained Multi-Hop Reasoning for Scientific Literature Synthesis
Sauhard Dubey
TL;DR
SciNets reframes cross-domain scientific synthesis as graph-constrained multi-hop reasoning over query-local concept graphs, enabling structured exploration of mechanistic links across fragmented literature. It separates symbolic reasoning from language realization and evaluates systems using a behavioral framework that emphasizes reasoning depth, diversity, and grounding stability rather than correctness. The study finds a consistent trade-off: deeper symbolic reasoning yields richer but less stable grounding, while shortest-path strategies produce reliable, well-grounded explanations. By introducing trace-based evaluation and releasing full reasoning traces, SciNets provides auditable, interpretable artifacts to support human-in-the-loop scientific inquiry and AI-assisted exploration across domains.
Abstract
Cross-domain scientific synthesis requires connecting mechanistic explanations across fragmented literature, a capability that remains challenging for both retrieval-based systems and unconstrained language models. While recent work has applied large language models to scientific summarization and question answering, these approaches provide limited control over reasoning depth and structural grounding. We frame mechanistic synthesis as a graph-constrained multi-hop reasoning problem over literature-derived concept graphs. Given a scientific query and a compact, query-local corpus, SciNets constructs a directed concept graph and synthesizes mechanistic explanations by identifying multi-hop reasoning paths that connect concepts that rarely co-occur within individual papers. We systematically compare shortest-path reasoning, k-shortest paths with diversity constraints, stochastic random walks, and a retrieval-augmented language model baseline. Rather than evaluating correctness, which is often indeterminate when synthesizing connections across distributed sources, we introduce a behavioral framework that measures symbolic reasoning depth, mechanistic diversity, and grounding stability. Across machine learning, biology, and climate science tasks, explicit graph constraints enable controllable multi-hop reasoning while revealing a consistent trade-off: deeper and more diverse symbolic reasoning increases grounding instability, whereas shortest-path reasoning remains highly stable but structurally conservative. These findings provide a systematic behavioral characterization of the limits and capabilities of current graph-LLM integration for scientific synthesis.