Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
110 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
44 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

BookWorld: From Novels to Interactive Agent Societies for Creative Story Generation (2504.14538v1)

Published 20 Apr 2025 in cs.CL

Abstract: Recent advances in LLMs have enabled social simulation through multi-agent systems. Prior efforts focus on agent societies created from scratch, assigning agents with newly defined personas. However, simulating established fictional worlds and characters remain largely underexplored, despite its significant practical value. In this paper, we introduce BookWorld, a comprehensive system for constructing and simulating book-based multi-agent societies. BookWorld's design covers comprehensive real-world intricacies, including diverse and dynamic characters, fictional worldviews, geographical constraints and changes, e.t.c. BookWorld enables diverse applications including story generation, interactive games and social simulation, offering novel ways to extend and explore beloved fictional works. Through extensive experiments, we demonstrate that BookWorld generates creative, high-quality stories while maintaining fidelity to the source books, surpassing previous methods with a win rate of 75.36%. The code of this paper can be found at the project page: https://bookworld2025.github.io/.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (6)
  1. Yiting Ran (2 papers)
  2. Xintao Wang (132 papers)
  3. Tian Qiu (73 papers)
  4. Jiaqing Liang (62 papers)
  5. Yanghua Xiao (151 papers)
  6. Deqing Yang (55 papers)

Summary

Analyzing the "BookWorld" Approach: Simulating Literary Worlds through Multi-Agent Systems

The paper "BookWorld: From Novels to Interactive Agent Societies for Creative Story Generation" introduces an intriguing framework for generating interactive story environments based on established literary works. Authored by researchers from Fudan University, this paper explores the possibility of transforming static narratives from novels into dynamic multi-agent societies that allow for the simulation of social interactions and story creations. The research leverages advanced LLMs to enhance the narrative generation process, providing a bridge from traditional literature to modern artificial intelligence applications.

The primary contribution of the paper lies in its BookWorld system, which uniquely positions itself within the field of LLM-enhanced storytelling by focusing on reproducing the complexities and dynamics of established fictional worlds. Unlike previous methodologies that primarily focus on generating agent societies from scratch, BookWorld emphasizes extrapolating from pre-existing fictional contexts, offering a robust framework for further exploration in narrative AI.

Core Methodology

BookWorld employs a two-tiered system consisting of role agents and a world agent. The role agents simulate individual characters, each possessing its own set of static and dynamic attributes extracted from the source literary work. These include fundamental characteristics and evolving memories that the system updates as the story progresses. In contrast, the world agent orchestrates the overall environment, ensuring continuity and logical adherence to the original fictional settings. This approach allows for a seamless integration of character-driven storylines and environmental interactions, offering a more immersive experience for users.

The system's simulation pipeline is initiated through detailed data preparation. Character profiles and the fictional world's worldview are meticulously extracted and structured, allowing for faithful reproduction of the original books’ settings. This structured data forms the basis of the BookWorld engine, enabling role agents to generate narratives that adhere to both character fidelity and story immersion.

Key Findings and Evaluations

The paper presents empirical results demonstrating the efficacy of BookWorld in generating coherent and high-quality narratives. Evaluations utilize pairwise comparisons between outputs from BookWorld and prior methods, such as direct generation and the HoLLMwood framework. Notably, BookWorld consistently outperforms baseline models in generating narratives across several dimensions: anthropomorphism, character fidelity, immersion, and more.

A rigorous ablation paper further highlights the significance of core BookWorld functions, revealing substantial declines in narrative quality when key features such as environment responses or comprehensive settings are disabled. These results emphasize the importance of incorporating detailed environmental and character-specific knowledge into the simulation process, a feature that BookWorld effectively implements.

Implications and Future Directions

The development of BookWorld marks a notable advancement in the harnessing of multi-agent systems for creative storytelling, setting a precedent for further research in the simulation of complex narrative environments. Its implications extend across various domains, such as digital storytelling, interactive gaming, and educational tools that enhance literary engagement.

The paper acknowledges certain limitations, particularly the system's reliance on simplified interactions that may not fully capture the intricate decision-making processes of characters in more elaborate scenarios. Future research may address these limitations by integrating more nuanced decision-making frameworks and expanding the system’s capabilities to simulate diverse literary genres.

Conclusion

The "BookWorld" paper contributes substantial advancements to the field of AI-driven narrative generation, particularly in its innovative use of LLMs to simulate and extend literary worlds into interactive environments. This research provides a foundational framework for future developments in narrative AI, highlighting the potential to create rich, dynamic experiences from existing literary works. It encourages exploration into more sophisticated simulations and broader applications, impacting how AI technologies can reinvigorate traditional narratives with modern digital interactivity.

Github Logo Streamline Icon: https://streamlinehq.com

GitHub

Youtube Logo Streamline Icon: https://streamlinehq.com