Daejeon, July 27–31
Chinese ancient book edition studies investigate how different editions were created, transmitted, and transformed over time (Chen 2006; Li et al. 2021). To reconstruct an edition’s evolution, researchers must determine its editing and publication time—information that must be inferred from bibliographic features such as layout, character taboos, and publisher marks. Another crucial task is identifying inheritance relationships between editions, which requires comparative analysis of textual content and physical features (Huang 2009). Because much of this information is uncertain, edition evolution is fundamentally a process of scholarly interpretation rather than straightforward fact retrieval.
Such interpretive practice has traditionally relied on close reading and manual extraction of evidence (Wei / Wang 1983). To enhance the accessibility and reusability of this knowledge, digital technologies can be introduced. However, although many digital humanities initiatives have sought to standardize knowledge (Romary 2010; Koster 2017), such as Six Degrees of Francis Bacon (Warren et al. 2016) and CBDB (Harvard University / Academia Sinica / Peking University 2021), most treat it as static and objective, insufficiently recognizing the interpretive processes that produce it.
This study therefore situates itself at the intersection of traditional textual scholarship and digital humanities, aiming to design a Linked Open Data model that can record both edition metadata and the interpretive processes that generate such metadata, while also enabling intuitive visualization of edition evolution. We use Shishuo Xinyu, a 5th-century collection of Chinese anecdotes compiled by Liu Yiqing as an example (Yu 1983). As a widely circulated work, it gave rise to numerous editions in the Sinosphere, and numerous scholars have contributed to the research of its edition evolution (Wang 1992; Zhang 2001).
This study asks: how can uncertain and interpretive metadata relating to the evolution of ancient book editions be structured while accommodating its uncertainty and subjectivity?
Several studies have attempted to construct data models for the evolution of Chinese ancient book editions (Ning 2022; Kimura 2022) , but most of them focus solely on structuring the metadata, while neglecting how this data is created. The neglect of the knowledge production process easily gives rise to the misconception that edition information is certain and objective, a view inconsistent with the interpretive nature of edition information.
Linked Open Data (LOD) is a paradigm of publishing structured data on the web enabling interlinking and reuse across domains. It structures data with URIs and vocabularies based on semantic web standards (Bauer / Kaltenböck 2011). Many researchers have sought to structure ancient Chinese books with LOD technologies, some are based on textual contents, such as Cong and Takaku (Cong / Takaku 2022), who developed a LOD model for Tang poems included in Japanese textbooks and employed Text Encoding Initiative (TEI Consortium 2025) to implement content annotations, enabling identification and comparison of textual variants; some focus on the original images of ancient books, such as Nakamura, Narita and Nagai (Nakamura et al. 2018), who developed a support system based on LOD and International Image Interoperability Framework (IIIF) (IIIF Editorial Committee 2015) for comparing the IIIF images from multiple editions of calligraphic models. Most of these studies do not consider the uncertainty and subjectivity inherent in humanistic knowledge. In contrast, Chen (Chen 2022) developed a more comprehensive model for the Han Slips from Edsen-gol, where CIDOC Conceptual Reference Model (CIDOC-CRM Special Interest Group 2024) was employed as a reference framework to model the interpretive practice through which scholars create annotations on the slips. This study draws on models of interpretive practices through which metadata of ancient books are produced.
This study develops a three-stage methodology: designing a LOD-based data model, constructing a corresponding dataset for Shishuo Xinyu, and implementing a visualization system to present edition evolution.
This study constructs an event-based data model which can organize information around events, rather than static object properties. In this approach, relationships are mediated by events rather than static attributes (Dijkshoorn et al. 2018). The model is grounded in the CIDOC Conceptual Reference Model (prefix: crm) and complemented by the Web Annotation Data Model (prefix: oa) (W3C Web Annotation Working Group 2017), Citation Typing Ontology (Peroni / Shotton 2012), DCMI Terms (DCMI Usage Board 2020), and a newly defined schema named Edition Evolution (prefix: ev).
The model consists of three integrated components: information sources, metadata, and time information.
In the information source component, interpretive statements derived from either external scholarly literature or internal evidence from book pages are modeled as ev:Statement. Their creation is represented as crm:E65 Creation events that explicitly record the agent responsible for the interpretation, the motivation behind it, and the specific information source from which it was inferred. Internal evidence, such as seals, characters, and page features, is described through IIIF-linked oa:Annotation entities, enabling precise identification of image regions supporting interpretation. In this way, the model captures not only the extracted information but also the evidential basis and interpretive context that give rise to it.
The metadata component extends crm:E13 Attribute Assignment class to represent the assignment of uncertain or interpretive metadata to editions. Rather than linking metadata directly to editions as static properties, the model represents each assignment as an event. This event records the agent, time, and supporting statements. This event-based representation accommodates the coexistence of multiple interpretations by different scholars, thereby preserving the epistemic structure of scholarly knowledge. Inheritance relationships between editions are expressed by linking editing and publication events to other editions through properties such as ev:basedOn and ev:republishedFrom, ensuring that genealogical relationships are represented not as static attributes but as interpretive outcomes of historically situated events.
The structuring of ancient time expressions is achieved using crm:E52 Time-span and its subclass ev:AncientBookTime. This design enables the simultaneous preservation of original reign-year expressions and the assignment of their standardized temporal interpretations. Multiple interpretations of the same time expression can coexist, reflecting scholarly disagreement or the presence of ambiguous evidence. These structured timespans are linked to editing or publication events via crm:P4 has time-span, making it possible to visualize both temporal uncertainty and the diversity of scholarly views. This enables both philological authenticity and machine-readable reasoning.
Based on the proposed data model, a dataset of 568 RDF triples was manually constructed for 19 editions, 11 facsimiles, 9 references, and related entities.
To enable exploration of structured information, a web-based visualization system was developed using Bokeh. The system integrates network and timeline visualizations across three dashboards: one showing the overall genealogy and temporal evolution of editions, another showing the provenance of metadata and relationships between statements and sources, and a third presenting detailed time information and associated internal evidence.
The evaluation employed three use-case scenarios derived from common scholarly inquiries and operationalized them through SPARQL queries executed on the constructed dataset. This method verifies whether the model can represent uncertain metadata, express non-standard time information, and retrieve interpretively relevant relationships.
Finally, the results demonstrate that the model effectively encodes multiple competing interpretations of editing or publication time, preserves both original and standardized temporal expressions, and accurately represents genealogical dependencies between editions. And the visualization system successfully presents these entities and relationships.
The findings show that the event-based approach effectively addresses the research question. It formalizes metadata and records the inferential pathways that lead to specific scholarly claims, thus capturing uncertainty and subjectivity.
However, several limitations remain. The data model lacks a structured terminology system for describing technical bibliographic concepts in Chinese philology, and its current temporal representation does not incorporate probabilistic or fuzzy temporal reasoning, which would better reflect the ambiguity present in many historical sources.
Besides, the dataset is small due to the labor-intensive nature of manual data extraction, limiting the scope of evaluation; the visualization system does not yet provide seamless integration of IIIF page viewing within all dashboards, hindering direct switching between metadata graphs and image-based evidence.
Additionally, although the model provides a mechanism for representing conflicting information from different scholars, further investigation is needed into how the reliability of such information can be evaluated at the application stage.
Furthermore, the usability of the proposed model has not yet been empirically evaluated. In future work, given the complexity of the event-based modeling approach, it is necessary to conduct usability testing with domain experts to assess whether the model can be effectively applied in real research scenarios.
This research adopts an event-based LOD model to structure the interpretive metadata concerning the evolution of ancient book editions. It provides a foundation for future work involving expanding the dataset, integrating richer terminological structures, enabling more sophisticated temporal reasoning, incorporating IIIF images more deeply into the system’s interface, and conducting user-centered evaluations.