Characteristic Analysis: Decoding the Social DNA of Fictional Universes

TL;DR

Is the social structure of Harry Potter fundamentally different from War and Peace? This paper argues that fictional social networks (SNF) mirror real-world social atoms. By mapping character interactions to graph nodes and edges, the researchers provide a mathematical lens to classify novels as "heroic" or "epic" based purely on topology, revealing that literature follows the same power-law distributions as our real-life interactions.

Background & Motivation: Beyond Manual Reading

Linguistic analysis of literature has historically been a manual, time-consuming process. While digital humanities have progressed toward word-level frequency analysis, they often miss the "social architecture" of the story. The authors propose that we don't need to understand every subtle word to understand the story's soul; instead, we can treat characters as vertices in a complex system. By focusing on Inter-relationship Strength, they bridge the gap between graph theory and literary criticism.

Methodology: High-Resolution Character Interaction

The core innovation lies in how character relationships are quantified without deep semantic parsing.

1. The Interaction Metric

The authors define interaction based on Statement Distance. If two characters appear in the same or adjacent sentences, their relationship is strong. The relationship weight is calculated as: $Inter(c_i,c_j)=\sum {\alpha }^{|l-m|}$ Where $\alpha$ (set to 0.5) ensures that the further apart two characters appear in the text, the weaker their link becomes. This is then converted into a distance metric using a logarithmic scale to represent "social distance."

2. Character Importance (k-leaf Algorithm)

To distinguish protagonists from peripheral characters, the authors use a recursive pruning technique on a Minimum Spanning Tree (MST). By iteratively removing leaf nodes ($k$-leaves), they find that protagonists consistently survive to the highest depth ($k$), providing a numerical ranking of character importance.

Figure 1: The k-leaf algorithm. Higher k-values correspond to more central, important characters.

Experiments & Deep Insights

The researchers tested their model on 20+ novels, including Three Kingdoms (912 characters) and Harry Potter (287 characters).

Heroic vs. Epic Structures

The study revealed a fascinating topological divide:

• Heroic Novels (e.g., Harry Potter): Show a "hub-and-spoke" model where almost all interactions are concentrated around a single hero. The distance distribution is tight, and the network is resilient to edge cutting.
• Epic Novels (e.g., War and Peace, Three Kingdoms): Feature multiple clusters of protagonists. The "distance" between characters is larger on average, representing a more fragmented, multifaceted social world.

Figure 3: Distance distributions revealing the level of focus on a protagonist.

The Power-Law Invariant

Regardless of the language (Korean or English) or the era of publication, the distribution of character co-occurrences followed a Power-Law. This suggests that authors, perhaps subconsciously, construct fictional worlds that obey the same mathematical constraints as real human communities—a phenomenon known as "Structural Invariance."

Critical Analysis & Future Outlook

Takeaway: This work proves that the "skeleton" of a novel can be extracted using syntactic distance alone. You don't need a PhD in literature to identify that ZhuGe Liang is the anchor of Three Kingdoms; the graph topology says it all.

Limitations: The current model struggles with Aliases and Pronouns. By only looking for full names, the system might miss subtle interactions (e.g., a character referred to only as "he" for three chapters).

Future Work: The integration of Named Entity Recognition (NER) and Coreference Resolution will be the next frontier. If we can map pronouns to characters with high precision, these social graphs will become even more granular, potentially allowing for the automated "fingerprinting" of an author's narrative style.