Narrative Knowledge Weaver: Narrative-Centric Retrieval-Augmented Reasoning for Long-Form Text Understanding

Researchers introduce Narrative Knowledge Weaver (NKW), a framework that improves AI's ability to answer questions about long-form narratives by integrating textual evidence, graph structures, and entity profiles to better understand story progression and character dynamics. The system outperforms existing retrieval methods on screenplay-based benchmarks while maintaining competitive performance on passage-focused tasks.

Narrative Knowledge Weaver addresses a fundamental limitation in how AI systems process long-form text. Existing retrieval-augmented generation methods treat stories as collections of disconnected passages, missing the interconnected nature of narrative—how character motivations evolve, temporal sequences matter, and causal chains link events across pages. NKW's innovation lies in encoding these narrative dimensions explicitly through atomic facts, entity profiles, and storyline structures that capture how evidence functions within a story world rather than in isolation.

This advancement emerges from the broader AI trend toward more sophisticated reasoning architectures. As language models scale, researchers increasingly recognize that raw retrieval and generation aren't sufficient for complex reasoning tasks. Methods like graph augmentation and multi-hop reasoning have shown promise, but NKW demonstrates that story-specific ontologies—tracking actor states, temporal positions, and narrative scope—unlock capabilities these general approaches miss. The framework's performance gains on screenplay and novel-based benchmarks suggest narrative structure is a core component of human comprehension that pure statistical models struggle to replicate.

For the AI development community, NKW indicates growing infrastructure around domain-specific retrieval systems. This pattern mirrors enterprise AI adoption, where off-the-shelf models require significant customization for specialized domains. Educational technology, content analysis platforms, and narrative-heavy applications could benefit from these techniques. The framework's emphasis on auditing constraints—polarity, scope, temporal position—also suggests progress toward more interpretable AI reasoning, potentially valuable for applications requiring explainability.

• →NKW integrates textual evidence, graph structures, and narrative components to improve long-form story understanding beyond traditional retrieval methods.
• →The framework excels at screenplay-level QA by explicitly encoding how character states, temporal sequences, and causal chains evolve through narratives.
• →Complementary benefits across character, scene, temporal, and causal reasoning suggest narrative ontologies are fundamental to story comprehension tasks.
• →Results indicate specialized retrieval architectures outperform general approaches for domain-specific reasoning in structured narrative domains.
• →The work advances interpretable AI by auditing constraints like polarity and scope, moving toward more explainable reasoning systems.