narrative character analysis

Narrative structure in literary analysis remains challenging to quantify and model systematically. Method: This study proposes an LLM-driven automated narrative parsing and visualization framework. It introduces an end-to-end LLM-based information extraction pipeline to precisely identify fine-grained interactions among characters, locations, and themes in novels and scripts. It innovatively redefines story-line visualization by designing an interactive system supporting dynamic, cross-level (macro-plot to micro-event) and multi-scale (temporal, spatial, semantic) exploration, augmented with human-in-the-loop mechanisms to mitigate LLM uncertainty. Contribution/Results: Evaluated on 36 canonical literary works, the method significantly improves both efficiency and accuracy in generating structured narrative data. It uncovers latent relational patterns and structural motifs overlooked by traditional close reading, thereby establishing a scalable, interpretable analytical infrastructure for digital humanities research.

This study addresses the longstanding reliance on subjective judgment in narrative quality assessment by introducing a computational framework grounded in 33 quantifiable linguistic features spanning lexical, syntactic, and semantic dimensions. For the first time, this work systematically applies multidimensional quantitative stylometric indicators to the automatic evaluation of narrative quality. Leveraging natural language processing, clustering analysis, and similarity matrix construction, the proposed model achieves near-perfect discrimination between texts authored by professional editors and self-published writers. Furthermore, it significantly outperforms existing evaluation metrics on a manually annotated dataset, thereby overcoming the limitations inherent in traditional story-level assessment approaches.

Existing narrative research suffers from small-scale, coarse-grained, and poorly generalizable character typologies, lacking a reliable benchmark to evaluate models’ capacity for understanding dynamic character development. Method: We propose a fine-grained character attribute attribution task and introduce CHATTER—the first large-scale, manually annotated dataset of character attributes in film scripts (660 films, 2,998 characters, 12,967 attributes, 88,124 character–attribute pairs)—built via structured script parsing, cross-character–attribute semantic alignment, and rigorous quality control. A high-quality subset, CHATTEREVAL, validated through multi-round human annotation and inter-annotator agreement assessment, is released as a new evaluation benchmark. Contribution/Results: Empirical evaluation reveals substantial limitations of state-of-the-art language models in reasoning about temporally evolving character attributes. CHATTER provides a scalable, reproducible, and narratively grounded evaluation infrastructure for narrative understanding and long-context modeling.

This study investigates the temporal evolution of religious belief and practice in Holocaust survivors’ oral testimonies. Employing natural language processing and first-person narrative analysis, we develop a dual-dimensional “belief–practice” thematic trajectory model, integrating temporal modeling and cross-individual clustering to identify shared evolutionary patterns. Our analysis reveals, for the first time, a universal structural pattern: religious belief exhibits long-term stabilization, whereas religious practice displays marked oscillation—reflecting the tension between internal conviction and external adaptation under extreme duress. This quantified narrative framework advances historical trauma research by providing a reproducible analytical paradigm; contributes empirical grounding to sociological theories of belief dynamics; and extends digital humanities methodologies in oral history. The approach bridges computational linguistics, narrative theory, and historical sociology, enabling rigorous, scalable analysis of subjective experience in archival testimony. (149 words)

Narrative visualization of complex data remains challenging in terms of interpretability and engagement. Method: We conduct a systematic literature review of 66 papers to propose the first end-to-end, four-stage reference model—Analysis, Narrative, Visualization, and Interaction—and distill eight core tasks, including insight extraction and author assistance. We further introduce a unified technical framework integrating large language models, multimodal understanding and generation, data insight mining, and human-AI collaboration, and systematically evaluate performance boundaries and challenges across tasks. Contribution/Results: We construct a structured knowledge graph that clarifies technological applicability scopes and open research questions, delivering an actionable roadmap and evaluation guidelines for researchers and practitioners in narrative visualization powered by foundation models.

This study investigates the differences between stories generated by large language models (LLMs) and those authored by humans with respect to character diversity. Drawing on narrative theory, the authors propose a novel analytical framework comprising eight dimensions—such as stylization and coherence—and integrate automated classification techniques to conduct both quantitative and qualitative comparisons. The findings reveal that while LLMs approach human-level performance on certain character dimensions, they still exhibit significant gaps in diversity and depth. By systematically delineating the capabilities and limitations of LLMs in character construction, this work advances the theoretical and methodological foundations for evaluating narrative competence in artificial intelligence systems.

This work addresses the challenges of generating accurate character descriptions from novels, which include tracking evolving attributes, synthesizing dispersed evidence, and inferring implicit information. The authors propose a two-stage framework that decouples reasoning from generation: first, a question-answering–guided structured reasoning mechanism produces a faithful reasoning trace grounded in the source text; then, a generative model produces the final character description based on this trace. This approach effectively circumvents interference from the built-in reasoning of large language models and is compatible with both long-context processing and chunked input strategies. Evaluated on the BookWorm and CroSS datasets, the method significantly outperforms strong baselines in terms of faithfulness, informativeness, and grounding in textual evidence.

This study addresses a critical gap in computational narratology by systematically examining how character functions vary across Frye’s four mythoi—comedy, romance, tragedy, and satire—a dimension largely overlooked in prior work focused primarily on narrative patterns. Integrating Jungian archetypal theory with Frye’s generic framework, the paper proposes a computable model comprising four universal character functions and their sixteen genre-specific instantiations. The model is rigorously evaluated using six state-of-the-art large language models, employing balanced accuracy and Fleiss’ κ inter-rater reliability on carefully constructed positive–negative sample pairs. Results demonstrate an average balanced accuracy of 82.5% (κ = 0.600), confirming that character functions exhibit structured, genre-dependent variation. Notably, the analysis reveals systematic role distributions in romance and deliberate subversions of archetypes in satire, underscoring the model’s capacity to capture nuanced narrative dynamics across genres.

This study addresses the limitations of traditional character modeling in literary analysis, which overemphasizes appearance frequency while neglecting crucial narrative dimensions such as discourse about characters by others and distinctions between narrator and character voices. Drawing on Woloch’s theory of “the one vs. the many,” the authors propose a six-dimensional structural model for character representation. Integrating large language models with a task-specific Transformer architecture, this approach enables a multidimensional computational characterization of fictional characters. It moves beyond frequency-centric paradigms by offering the first quantitative operationalization of implicit dimensions like “discourse by others.” Empirical validation on a corpus of 19th-century British realist novels not only confirms key theoretical assumptions but also uncovers nuanced relationships between character centrality and gender dynamics.

Existing narrative visualization approaches are often limited to timelines depicting character–location co-occurrences, struggling to represent deeper narrative structures such as focalization and causality. This work introduces the literary concept of focalization into narrative visualization in a systematic manner, proposing a novel method to model how characters perceive, participate in, observe, or narrate events. Accompanying this conceptual advance is an interactive tool that enables seamless integration of textual content and visual representation. By synthesizing narratological modeling, information visualization design, and interactive techniques, the approach was qualitatively evaluated by four writers and scholars, who found it effective for drafting reflection and literary analysis. The method significantly enhances understanding of narrative perspective differences, offering creators and researchers an analytical dimension absent in traditional approaches.