JSAI2025

Presentation information

Poster Session

Poster session » Poster Session

[1Win4] Poster session 1

Tue. May 27, 2025 3:30 PM - 5:30 PM Room W (Event hall D-E)

[1Win4-48] Automatic Story Generation Using Large Language Models with Foreshadowing Creation and Plot Reconstruction

〇Narumi Uno1, Yoshinobu Kano1 (1.Shizuoka University)

Keywords:Novel, Large Language Model, Automatic Generation

The advancement of large language models (LLMs) has brought increasing attention to automatic novel generation.
However, LLMs tend to generate common and predictable choices, often resulting in mundane stories.
This study focuses on "unpredictable plot developments" and "foreshadowing resolution" and proposes a prompt design and generation method to incorporate these elements.
Our approach adopts a two-stage structure that alternates between plot generation and text generation.
In plot generation, we enhance unpredictability by adding "unexpected developments" to the initial plot.
In text generation, we explicitly introduce foreshadowing and ensuring its resolution. Additionally, after text generation, the plot is reconstructed to maintain coherence and improve the accuracy of foreshadowing resolution.
We generated plots across various genres and conducted human evaluations.
The results demonstrated that our proposed method significantly enhances unpredictability compared to conventional approaches, making the story more engaging for readers. Furthermore, the combination of explicit foreshadowing and plot reconstruction improved the resolution rate, increased the overall completeness of the story, and enabled the generation and resolution of more compelling foreshadowing elements.

Please log in with your participant account.
» Participant Log In