Computational Modeling and Analysis of Narrative Structures in Tolkien's Middle-earth Using AI and Graph-Based Methods A Data-Driven Approach to Character Networks, Semantic Structures, and Story Evolution

For ages, in the world of studying stories, figuring out how a story’s characters grow, how its main ideas unfold, and how it’s put together has been hugely important. In the past, this meant giving a really detailed, thoughtful reading of the story...and that’s a method that, though it gets you deep into the story, is bound to be based on someone’s opinion and is hard to do with lots and lots of stories. But now that digital humanities are using computers, we have fresh chances to look at the way stories are built, but this time, by looking at the information itself [5], [19].

We’ve built a way to use computers to understand how stories are put together in The Lord of the Rings and The Hobbit, and maybe even The Silmarillion if you like.  Using all the clever developments in Natural Language Processing (NLP), and specifically powerful ‘transformer’ models like BERT [2] and how words get meaning from their context, we’re pulling out who is in the stories, how they relate to each other, and the big ideas running through them [1]. Then, we’re turning all of that into a sort of map of connections, a graphical structure which lets us use tools from network analysis and the study of complicated systems to explore the books…and the research of Barabasi [14] and others [17] gives us a base for doing that

We build ever-changing maps of who the characters in a story are, and how they’re connected, by first identifying the characters themselves (Named Entity Recognition), then figuring out what they mean (semantic embedding) and how often they appear together (co-occurrence analysis). Looking at the most important figures within these networks, and breaking the network down into groups, shows us the way the story is organized and how the relationships between people change as events unfold. Simultaneously, we look at the ideas within the story, grouping similar subjects (topic modeling and semantic clustering) to understand how themes grow and what the story starts to emphasize at different points [7], [19].

Preliminary findings indicate that Tolkien’s legendarium exhibits a highly modular narrative structure, characterized by distinct yet interconnected sub-networks corresponding to major story arcs. Central characters such as Frodo, Aragorn, and Gandalf demonstrate varying degrees of influence over time, reflecting the dynamic redistribution of narrative focus. Additionally, semantic analysis reveals a progression of dominant themes, including power, sacrifice, and fellowship, aligning with key narrative transitions.

The primary contributions of this work are threefold: (1) the development of a unified computational framework combining NLP and graph-based methods for literary analysis, (2) the quantitative characterization of narrative complexity and character interaction patterns in Tolkien’s works, and (3) new insights into the evolution of narrative structure through temporal and semantic modeling. These findings demonstrate the potential of AI-driven approaches to complement and extend traditional literary scholarship, offering scalable and reproducible methods for analyzing complex narrative systems.

Keywords: Computational Narrative Analysis; Natural Language Processing (NLP); Graph Theory; Character Networks; Semantic Modeling; Tolkien’s Middle-earth; Narrative Evolution