GraphLit: Learning Text-Enriched Dynamic Character Network Representations for Literary Study

2026-05-27 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Advanced, medium

Summary

GraphLit is a novel self-supervised learning framework designed for literary study, which learns rich representations of long novels by integrating character interactions with their textual contexts. It introduces Dynamic Heterogeneous Character Networks (DHCNs), a method to organize novels into temporally localized heterogeneous graphs. Researchers extracted approximately 20,000 DHCNs from Project Gutenberg to train GraphLit using a masked graph autoencoder objective. This framework significantly improves performance across 12 character-related tasks, particularly those demanding contextual understanding, surpassing both text-only and graph-only baseline models. Furthermore, GraphLit and DHCNs demonstrate applicability in literary analysis, specifically for examining the relationship between narrative non-linearity and dynamic social features within texts.

Key takeaway

For NLP researchers or digital humanists analyzing complex narrative structures, traditional graph-only or text-only methods may overlook critical textual context. You should consider Dynamic Heterogeneous Character Networks (DHCNs) and the GraphLit framework to integrate character interactions with their surrounding text. This approach provides richer, context-aware representations, enabling deeper insights into dynamic social features and narrative non-linearity in literary texts.

Key insights

GraphLit integrates dynamic character networks with textual context via self-supervised learning to enhance literary analysis.

Principles

• Textual context enriches character interaction graphs.
• Dynamic graphs capture temporal evolution in narratives.
• Self-supervised learning yields rich literary representations.

Method

Novels are organized into Dynamic Heterogeneous Character Networks (DHCNs) as temporally localized graphs. GraphLit then applies a masked graph autoencoder objective on these DHCNs for self-supervised representation learning.

In practice

• Analyze narrative non-linearity in novels.
• Study dynamic social features within texts.
• Improve character-related task performance.

Topics

• GraphLit
• Dynamic Heterogeneous Character Networks
• Literary Analysis
• Self-supervised Learning
• Character Networks
• Project Gutenberg

Code references

• oucailab/SDANet

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Planning Beyond Text: Graph-based Reasoning for Complex Narrative Generation — Graph-based narrative planning significantly enhances LLM-generated story coherence and character development over text-based methods.
• From Global to Local: Learning Context-Aware Graph Representations for Document Classification and Summarization — Sliding-window attention for graph construction improves document representation efficiency and performance in NLP tasks.
• Knowledge Graph and Hypergraph Transformers with Repository-Attention and Journey-Based Role Transport — A novel architecture separates structured knowledge from language in transformers using a key-value repository and journey-based role transport.
• HOLONS: A New Hope — Context graphs, structured holonically, enable robust, queryable representations of dynamic information without requiring a triple store.
• GS-Quant: Granular Semantic and Generative Structural Quantization for Knowledge Graph Completion — GS-Quant aligns graph embeddings with LLM tokens using semantically coherent, hierarchically structured discrete codes.
• Deconstructing Sentences: Context Graphs and Reification — Reification in knowledge graphs enhances natural language representation by attaching properties directly to relationships.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.