Knowledge graphs can turn connected events into narratives that preserve the logic of what happened.
From Nodes to Narratives (Text2Story’24)
Generating a story from a knowledge graph is not only about linking people, places, and dates. It also requires capturing how events shape one another.
In this paper, we enrich event-centric knowledge graphs with fine-grained relations such as causing, preventing, enabling, and intending to cause. We then use these richer graphs to generate more informative narratives.
| Human evaluation | Combined model | Base model |
|---|---|---|
| More adequate narratives | 58.3% wins | 8.3% wins |
| More fluent narratives | 33.3% wins | 16.7% wins |
The results show that adding precise event relations helps move from isolated graph nodes to narratives that better reflect the meaning and connections behind events.