The Shang–Zhou period, spanning approximately from 1600 BCE to 256 BCE, represents a crucial developmental stage of China’s bronze civilization. It encompasses the following phases:

(1) the Shang dynasty (ca. 1600–1046 BCE);

(2) the Western Zhou (1046–771 BCE);

(3) the Spring and Autumn (770–476 BCE) and Warring States periods (475–256 BCE).

The core dataset of this study is The Compendium of Ancient Chinese Weapons, compiled by Shen Rong and published in two volumes by Shanghai Lexicographical Publishing House in 2015. The work adopts an annotated-entry format, comprising 14 sections, over 4200 entries, and nearly 1.5 million characters. It is a comprehensive reference work on ancient Chinese weaponry.

The compendium explicates the basic terms, entities, and related concepts of Chinese weapons spanning millennia, systematically presenting the development of China’s ancient military equipment system. It documents morphological characteristics, functional uses, and historical contexts in detail, integrating extensive historical texts, archaeological data, and weapon studies. Each entry cites references in a standardized format, ensuring data diversity and reliability. This compendium thus provides a rich terminological and data foundation for the ontology construction of Chinese bronze weapons.

In addition to this core resource, a wide range of archaeological excavation reports and specialized research literature on bronze weapons were also consulted to enhance data coverage, reliability, and comprehensiveness.

The classification and naming of bronze weapons have generally followed two approaches:

(1) Functionalist systems, grounded in terminology and descriptive conventions from epigraphy and antiquarian studies. These systems typically divide weapons into offensive and defensive categories. For instance, Ma Chengyuan’s seminal work (1982; 1988) treats all weapons as a distinct class, further splitting them into close‑combat and ranged types. Du Naisong (1984) proposed a broader twelve-category scheme, listing Ge, swords, and crossbows as major types. Similar functional schemas were adopted by international scholars such as Umehara (1940), Lin (1972), and Karlgren (1930), who emphasized sharpness, form, and technological attributes.

(2) Morphology-based systems, emerging alongside modern archaeological excavations, emphasize formal characteristics—such as main blade shape, inner type, and hafting method—over textual references. Li Ji’s (1952) classification in Report on Bronze Artifacts Unearthed from Xiaotun identifies four main types: pointed, end-blade, side-blade, and double-edged weapons, with subtypes including hooked, piercing, and long-bladed forms. Although this morphological taxonomy was not widely institutionalized, it provides a valuable reference for formal structure modeling in ontological frameworks.

The development of functionalist and morphology-based classification systems has established an essential descriptive framework for bronze weapon studies. However, the absence of explicit, standardized definitions for categories and terms has led to semantic ambiguity, while the coexistence of divergent typological schemes across scholars and sites has hindered data harmonization. These limitations underscore the need for a unified, formalized terminology system that can reconcile historical classification traditions with the precision required for computational analysis.

The rise of Semantic Web technologies has introduced formal ontology languages—Web Ontology Language (OWL) /Resource Description Framework (RDF) —to cultural heritage. Foundational models such as SKOS (Miles and Bechhofer, 2009), CIDOC CRM (Doerr, 2005), and Europeana EDM (Doerr et al., 2010) provide broad frameworks for describing events, places, and terminology. In Chinese heritage, ontologies have been successfully applied to ceramics (Wei et al., 2022) and ritual bronzes (Wei and Chen, 2023b). However, no formal ontology has yet been developed to semantically represent bronze weaponry.

Recent research has extended ontology modeling to military history. Notable examples include the WarSampo project, which extends CIDOC CRM with event ontologies to model battles, participants, locations, weapons, and timelines in a semantically searchable knowledge graph (Hyvönen et al., 2016). In Chinese contexts, semantic modeling remains exploratory: existing projects primarily focus on military terminology or campaign chronologies, with limited attention to weapons themselves (Lu et al., 2024). Bronze Age armaments in particular remain underrepresented, and few ontologies bridge weapon entities with combat events, actors, and functional roles.

Beyond domain-specific applications, developments in Knowledge Organization (KO) offer important epistemological and methodological resources for cultural heritage ontology work. Hjørland (2013) argues that all knowledge organization systems embody implicit epistemic and ontological commitments; applied to artifact studies, this implies that classificatory choices—whether morphological, functional, or historical—reflect specific scholarly perspectives that should be made explicit in formal models. Mazzocchi (2018) further observes that formal ontologies, despite their logical rigor, inevitably encode cultural assumptions, reinforcing the importance of documenting the provenance, scope, and perspective of domain concepts. Incorporating these insights encourages heritage ontologies to represent not only canonical hierarchies but also the plurality of viewpoints that underpin archaeological typologies.

Methodologically, KO’s domain-analytic tradition stresses the importance of examining disciplinary discourse before concept formalization. This stance resonates with Bagchi’s (2021b) knowledge-intensive modeling pipeline, which emphasizes iterative refinement, conceptual transparency, and evidence-based category construction. Likewise, the problem of overlapping classificatory criteria highlighted in KO research parallels Bagchi and Das’s (2022) call for conceptual disentanglement in classificatory ontologies. Emerging research also indicates that human–LLM collaboration can support early-stage metadata modeling by producing candidate conceptual structures for expert review (Bagchi, 2025), pointing toward new methodological possibilities for complex heritage domains.

Thus, while Semantic Web technologies provide powerful means of knowledge representation, their application to bronze weapons remains relatively underexplored and limited in scope and granularity. Current cultural heritage ontologies either adopt coarse levels of granularity that obscure morphological distinctions or focus narrowly on events and terminology without modeling the internal complexity of weapons. This gap prevents effective linkage between morphological data and broader historical contexts. Therefore, constructing a fine-grained ontology of Shang–Zhou bronze weapons—prioritizing structural representation and standardized terminology—constitutes a critical step in extending the ontology lineage of cultural heritage. Such a framework can provide semantic foundations for future integration with event- and actor-based models, and ultimately support large-scale, heterogeneous data analysis in military and cultural heritage studies.

This study proposes an ontology construction framework that integrates semantic modeling of artifacts with the representation of historical contexts. Following the methodology of “term–concept–characteristic” and introducing the semantic chain of “weapon–actor–event”, the framework enables structured knowledge modeling and contextual knowledge organization of Shang–Zhou bronze weapons. The ontology adheres to the terminology standards ISO 1087 (ISO, 2019) and ISO 704 (ISO, 2009) as well as the OWL 2 DL modeling specification, and covers essential characteristic identification, terminology system construction, semantic structure design, and ontology integration. It supports fine-grained, computable modeling of weapon knowledge, ensuring semantic interoperability and ontology integration. The theoretical basis and operational workflow are illustrated in Fig. 1.

Fig. 1.

Ontology construction workflow for Shang–Zhou bronze weapons. The workflow includes seven steps: (1) scope and objective; (2) object and term base building; (3) essential characteristic identification; (4) defining concepts based on essential characteristics; (5) semantic chain modeling; (6) building ontology on Protégé; (7) ontology integration. CIDOC CRM, International Committee for Documentation Conceptual Reference Model; SKOS, Simple Knowledge Organization System.

4.1.3 Identifying Essential Characteristics

A locally closed-world assumption was adopted for characteristic identification: only expert-validated core structural attributes were considered, while contingent or unstable properties (e.g., color) were excluded to maintain logical rigor and class distinguishability. By analyzing essential differences between weapon classes, instances, and parts, this study extracted the essentialcharacteristics at each hierarchical level and identified key attribute dimensions. Classification criteria included functional, structural, and morphological dimensions.

Functional classification — commonly used in archaeology — partitions bronze weapons into: (1) close-combat weapons (e.g., Ge, spear, knife, sword); (2) ranged weapons (e.g., bow, crossbow, arrow); and (3) defensive equipment (e.g., armor, shield). Close combat weapons are subdivided into short-handled (knife, sword) and long-handled (ge, spear, long-knife). Ranged weapons are divided into projectile (bow, crossbow, arrow) and throwing types (javelin, slingshot). Defensive weapons are divided into armor and shields (Fig. 2).

Fig. 2.

Core class hierarchy of Shang–Zhou bronze weapons. The tree shows the functional–structural partition adopted under a locally closed-world assumption: top-level distinctions into close-combat, ranged and defensive classes are successively refined down to shaft type and firing method, respectively.

On the basis of this core classification, weapon parts were further identified and categorized through detailed analysis of functional and structural differences within each weapon type. For example, the key parts of a bronze Ge include main blade (yuan), inner part (nei), Hu, hole (chuan), side guard, upper and lower teeth, blade–inner junction, and haft. The main blade itself can be subdivided into main body, edge, tip, and spine (Fig. 3). Based on these parts, different morphological types of Ge can be defined—for example, horizontally extended main blade, triangular main Blade; rectangular straight inner part, or rounded curved inner part (Fig. 4). This approach was extended to other weapon categories, distinguishing unique and shared structural parts. For instance, close combat weapons commonly include edge, tip, haft, spine, and hole. Beyond components, decorative motifs and materials were also modeled as core classes.

Fig. 3.

Identification of the key parts of a Ge. This figure shows the key parts identification of Ge, including main blade, inner part, side guard, base, upper teeth, lower teeth and hole. The main blade is further divided into upper edge, lower edge, point and spine.

Fig. 4.

Sample characteristic Identification of Ge Instances. This figure shows the distinguishing characteristics of two Ge examples.

4.1.4 Defining Concepts Based on Essential Characteristics

The identified structural characteristics were combined into concepts, forming hierarchical models of weapon components. The process included: (1) terminology normalization: assigning unique and unambiguous terms to each component and characteristic; (2) essential characteristic identification: selecting defining characteristics in accordance with ISO 704 and ISO 1087 standards; (3) formalization: translating essential characteristics into DL (Description Logic) axioms to produce inferable class definitions; (4) annotation: recording non-essential but common characteristics as annotations for expert reference without logical entailment.

For example, the concept ChangHuSanChuanGuiYuan Ge (long Hu, three-holes Ge with Gui-shaped blade) is formally defined by the following characteristics:

ChangHuSanChuanGuiYuanGe =

{

hasPart some Long Strip Main Blade,

hasPart some Rectangular Straight Inner Part,

hasPart some Long Hu,

hasPart some Clamped-tang Haft,

hasPart some Side Guard of Ge,

hasPart some Lower Teeth of Ge,

hasPart some Upward Straight Upper Edge,

hasPart some Inward Curved Lower Edge,

hasPart some Gui Head Tip,

hasPart some Long Shaft,

hasPart some Linear Spine,

hasPart exactly 3 (

Hole

and (hasHoleShape value Rectangle Hole)

and (hasHolePosition value Long Strip Main Blade))

}

The formal composition of this concept, integrating the weapon classification hierarchy and the parts/characteristics hierarchy, is illustrated in Fig. 5.

Fig. 5.

Concept composition of the ChangHuSanChuanGuiYuan Ge based on essential characteristics. The concept is defined by the two core hierarchies of bronze weapons: parts/characteristics hierarchy (blue) and weapon classification hierarchy (yellow).

In natural language, this weapon can be described as:

A Ge with an long strip main blade bearing three rectangular holes and a Gui-shaped tip; a long Hu; a rectangular straight inner part connected by clamped inner part to a long shaft; an upward straight upper edge and inward curved lower edge; a main blade with linear ridge, side guard and lower teeth.

This illustrates the “term–concept–characteristic” framework and its formalization method. The framework is extensible: new weapon types (e.g., other forms of ge or spears) can be added simply by supplying their essential characteristic sets and instances, without altering the core class structure.

To meet the demands of semantic modeling, form expression, and cross-domain reuse of Shang–Zhou bronze weapons within the context of cultural heritage digitization, this study constructs a formal, hierarchical bronze weapon ontology. The ontology supports modeling at multiple granularities, covering the full semantic pathway from morphological description to semantic chain modeling, and is applicable to subsequent form analysis and knowledge graph construction. The ontology is constructed using the OWL 2 DL language and complies with formal modeling specifications under the Description Logic framework. The ontology includes 442 classes, 18 object properties, 12 data properties, and 4405 axioms. The core classes and properties are shown in Fig. 6.

Fig. 6.

Core classes and object properties in the ontology. Rectangles represent classes: blue rectangles indicate classes reused from CIDOC CRM, yellow rectangles indicate bronze-weapon ontology classes; arrows denote object properties.

4.2.4 Ontology Instances

To validate the modeling capacity and adaptability of the ontology, the bronze weapon Yanyu Ge from the late Warring States period of Zhao was selected as a representative example. An instance knowledge graph was constructed focusing on both morphological characteristics and historical context (see Fig. 7). This graph integrates the weapon’s physical characteristics, spatiotemporal information, and contextual attributes, specifically including:

$\bullet$ Structural Dimension: Components are linked via the object property hasPart, such as Long Hu, Long Strip Main Blade, Sharp Leaf-shaped Tip, Upward Straight Upper Edge, and Inward Curved Lower Edge, achieving fine-grained structural representation.

$\bullet$ Material and Craftsmanship: The material bronze and the crafting technique casting are defined through madeOf and craftedWith.

$\bullet$ Decoration and Function: The weapon’s inscription decoration and its close-range attack function are represented through decoratedBy and hasFunction.

$\bullet$ Spatiotemporal Localization: The place of manufacture and excavation are indicated by madeIn and foundIn as Shanxi Linxian and Yanyu, with temporal context linked to the late Warring States period via P4_has_time-span.

$\bullet$ Contextual Expansion: The ontology further links the weapon to its users and events: it was used by the Zhao Army (usedBy) in the Battle of Yanyu (usedIn), with the key Zhao general Zhao She (belongTo) associated with the relevant time (P4_has_time-span) and place (P7_took_place_at).

Fig. 7.

Instance knowledge graph of the Yanyu Ge. This figure illustrates the ontology’s capacity to integrate morphological, material, spatiotemporal and event data for a single weapon.

This example demonstrates the ontology’s semantic adaptability in modeling typical bronze weapons. It verifies the ontology’s effectiveness in integrating morphological, spatiotemporal, and contextual information, thereby showing its potential for artifact knowledge extraction, historical event association, and multi-source data integration. The case also provides a practical paradigm and instance foundation for subsequent knowledge graph construction.

Structural completeness and complexity control were assessed using the OntoMetrics online platform. Six commonly adopted ontology structural metrics were applied—Attribute Richness, Inheritance Richness, Relationship Richness, Class/Relation Ratio, Average Population, and Class Richness—to quantitatively analyze the ontology (Table 8).

The results reveal three key observations:

(1) Low attribute richness and average population indicate that individual classes contain relatively few attributes and instances. This reflects that the ontology is designed primarily as a structured classification framework rather than a fully populated dataset. Low attribute richness helps maintain simplicity and avoids excessive maintenance costs.

(2) Moderate inheritance and relationship richness show that each class has, on average, about 1.15 subclasses, and approximately 44.2% of the relations are non-hierarchical. This suggests that the ontology combines hierarchical depth with cross-class semantic links, balancing structural clarity with expressiveness.

(3) High class/relation ratio and class richness indicate strong connectivity, with most classes participating in relational structures. A class richness of 0.604 shows that over 60% of classes are instantiated, providing a solid foundation for applications. However, since instance population is not the ontology’s primary focus, some classes remain unused and may be populated in future work.

Overall, the six metrics confirm that the ontology achieves a balance between structural control, semantic expressiveness, and scalability.

To evaluate the practical semantic retrieval capacity of the ontology, competency questions (Table 1) were used as test cases. These questions were designed to assess the ontology’s effectiveness in terms of classification, terminology, and contextual modeling. Two representative competency questions are presented here (CQ1 and CQ4).

$\bullet$ CQ1 tests whether the ontology’s class hierarchy and terminology system effectively support type-level queries.

$\bullet$ CQ4 examines whether property relations (e.g., hasDecoration) can correctly link entities with descriptive attributes, thereby validating the “object–property–value” chain of semantic modeling.

The corresponding queries and results are shown in the following resources: Table 9 (SPARQL prefixes), Table 10 (CQ1 query), Fig. 8 (CQ1 results), Table 11 (CQ4 query), and Fig. 9 (CQ4 results).

Fig. 8.

The results of CQ1. The Chinese text in the figure is Types of Ranged Weapons, corresponding to the type-level query focus of CQ1 (query details in Table 10).

Fig. 9.

The results of CQ4. The Chinese text in the figure refers to knife names, which aligns with the attribute relation examined by CQ4 (query details in Table 11).

(1) Types of Ranged Weapons

(2) Decorations on Knife Instances

In addition to the two representative capability questions, all other queries also returned clear semantic responses, demonstrating the ontology’s strong abilities in type-level recognition, attribute-property association, and hierarchical reasoning. This capability evaluation confirms that the ontology performs well in archaeological knowledge organization, artifact attribute association, and semantic retrieval scenarios. It shows significant potential for supporting cultural heritage digital analysis and the construction of a Bronze Weapon Knowledge Graph.

Bronze sword characteristics were statistically analyzed across nine chronological phases (early/middle/late Shang, Zhou, Spring and Autumn, Warring States). Frequency and proportion of features were calculated, and trend charts were drawn (Fig. 10).

Fig. 10.

Example of morphological evolution of bronze sword parts (only four dimensions are shown here).

The figure shows the following conclusions at the sample level:

(1) Late Spring and Autumn Period: The number of samples significantly increased, and the Warring States period still had a large number of samples. Historically, from the early Western Zhou to the mid-Spring and Autumn periods, chariot warfare was the primary method of combat, with long-handled weapons like the Ge (polearm) and Spear being commonly used in chariot combat. By the late Spring and Autumn period, the focus shifted from chariot warfare to infantry/cavalry warfare, creating conditions for the spread of swords as personal hand-held close-combat weapons. In the Warring States period, large-scale wars and a mature military system broadened the use of swords.

(2) Morphological Changes: The turning point for morphological change occurred in the mid-Spring and Autumn period, where significant differences were found between the mainstream characteristics of the periods before and after. For example, after the mid-Spring and Autumn period, the proportion of swords with a guard increased significantly, with noticeable regional differences. Even earlier, there were examples of swords with a guard. From a craftsmanship perspective, the Spring and Autumn period saw the peak of bronze sword craftsmanship, with casting techniques (including composite casting) becoming more refined. Sword dimensions and forms changed significantly.

(3) Morphological Evolution Rate: The rate of morphological evolution showed significant phase differences. Before the mid-Spring and Autumn period, the proportion of various morphological characteristics fluctuated greatly, but after the mid-Spring and Autumn period until the Warring States period, the mainstream morphological characteristics became more stable. For example, the leaf-shaped short sword was common in the early Western Zhou period but quickly declined; after the mid-Spring and Autumn period, the orchid leaf-shaped sword dominated; and before the mid-Spring and Autumn period, the blade tip shapes were diverse, but after the mid-Spring and Autumn period, the disc-shaped tip became the dominant form. Overall, the characteristics showed a trend of “diversification leading to standardization”.

These patterns suggest (a) early immaturity of craftsmanship before standardized techniques were established, and (b) accelerated unification of forms after the mid-Spring and Autumn due to intensified warfare and trade networks.

Based on the annotated data guided by the bronze weapon ontology, this section further explores artifact clustering methods, with the goal of performing large-scale machine computations to cluster the morphological similarity of artifacts. To make the clustering results more significant, we reduce the influence of incidental factors on the clustering effect through several preprocessing steps. The clustering steps are as follows:

(1) Representative Artifacts: Duplicate types were reduced to one earliest instance to avoid overrepresentation.

(2) Similarity Matrix: Weapon characteristics were encoded as vectors, and cosine similarity was computed to form the similarity matrix.

(3) Network construction: Only strongest links were retained; similarity scores were used as edge weights, and data were imported into Gephi.

(4) Community detection: The Louvain algorithm was applied to group similar weapons (Fig. 11).

Fig. 11.

Similarity Clustering of Bronze Swords and Corresponding Legend. (a) Similarity clustering results for bronze swords. (b) Legend. The figure displays 17 clusters of bronze swords, with each cluster labeled by a numerical identifier corresponding to the English name listed in the legend. Representative examples include Willow-Leaf Shape Ring Tip Sword and King of Yan “Zhi” Sword. All terminology follows standardized English translations to ensure clarity for an international readership.

The clustering results yielded 17 clusters, with the largest cluster containing 16 instances and the smallest having 2. The modularity score was 0.863, indicating strong community cohesion. Traditional typology in ancient weaponry classification usually relies on the researcher’s experience, with mainstream views categorizing swords based on differences in characteristics such as blade, stem, ridge, guard, and tip. From the clustering results in this study, it is clear that swords with similar names are grouped into the same class. For example, swords named after their morphology (e.g., Cluster 11 Willow-Leaf Shape Single/Double Hole Short Swords) and those named after their origins (e.g., swords prefixed with “King of Yue” or “Changxing”) were automatically aggregated. This indicates that the clustering method implicitly contains the core logic of traditional classification while also revealing potential time and regional groupings, providing a reference for studying the evolution and spread of weapons.

Compared to traditional archaeological methods, the innovation of this study lies in using the ontology to specify the required characteristic types, encoding and computing fine-grained characteristics using universal numerical methods, thereby reducing the impact of subjective weighting on results. This offers an additional tool for traditional classification from a global, comprehensive perspective, enabling the division of large-scale artifacts into groups.

In summary, the ontology-driven bronze sword morphological computation method can reveal both macro-level evolutionary characteristics (e.g., turning points in the Spring and Autumn period) and micro-level similarity groupings between artifacts. Its advantages include reproducibility and cross-regional comparability, though its explanatory power remains limited by sample representativeness, characteristic encoding, and clustering algorithm selection. Future research could integrate traditional archaeology typology and metallurgy studies to form more robust conclusions.