Graph guided reasoning helps a small ai model answer long mystery questions
Beyond Linear Context: Graph-Guided Evidence Navigation for Long-Novel Reasoning with a Local 9B Language Model
Artificial IntelligenceComputation and Language
Summary
When small AI models read long mystery novels, they usually process text one word after another, which limits their understanding and memory. The authors tested if linking important clues using a knowledge graph, like a detective’s map of who is connected to whom, can help a small model reason better about the story. They found that guiding the model with these graphs improved its ability to answer questions about the novels compared to other reading methods. While the improvements are exploratory and need more testing, the work shows which parts of the graph and data are important for understanding long stories.
What this means in practice
- •For ai developers: Design small language model systems that use graph-based navigation to improve reasoning over long documents such as novels or reports.
- •For digital library teams: Enhance question-answering tools for large text collections by integrating knowledge graphs to help locate and connect relevant evidence across chapters or sections.
Authors
Wenji Fu
Abstract
Long-context models read a novel the way a person reads a printout: one token after another, in narrative order, with the whole history competing for a fixed budget of attention. A detective does not work that way. They sort what happened when, and they keep a map of who relates to whom, so a clue from chapter one can meet a question asked at the end of the book. We test whether a frozen knowledge graph can give a small local model that same freedom. Thirty detective novels and 234 multiple-choice questions are answered by one fixed qwen3.5:9b reader under nine conditions: five graph routes, a recent-window baseline, whole-book compression, ordinary vector retrieval, and a question-only control. The strongest graph route reaches 53.85% (126/234) against 46.15% for the recent window, 51.28% for compression, 51.71% for vector retrieval and 40.17% for question-only. On the subset that no model can answer without the book, the graph route reaches 42.86%. None of the fifteen graph-baseline contrasts survives Holm correction, so we present the result as exploratory evidence about a design. Two structural findings survive scrutiny better than the headline number: annotated evidence concentrates in the topological core of these graphs (2.35x enrichment, pooled), and the two graph-building pipelines differ so much in annotation coverage (16% versus 73% of clue paragraphs) that pooled accuracy alone would hide which bottleneck is being measured.