Global production processes are highly dependent on the resilience of global supply chains [1]. In order to measure the resilience of international trade flows various indicators from available information sources must be brought together as comprehensively as possible. A formal semantic description of international trade flows, taking into account existing standards, is suitable for such an integration of different data sources.
There are many global databases that provide insight into global supply chains [2]. For example, the World Input-Output Database (WIOD) [3], the Eora Global Supply Chain Database [4], the Global Trade Analysis Project (GTAP) [5] which use analytical models to study global supply chains, the database International Trade at Product Level (BACI), and Trade in Value Added (TiVA) databases [6].
To the best of our knowledge, these databases lack advanced analytical capabilities that use knowledge graph for semantic data integration and sharing across the various computational tools employed in resilience analytics.
Third International Workshop on Linked Data-driven Resilience Research (D2R2 '24) co-located with ESWC 2024, May 27th, 2024, Hersonissos, Greece Envelope Felix.Engel@tib.eu (F. Engel); Mark.Vanin@tib.eu (M. Vanin); Nenad.Krdzavac@tib.eu (N. Krdzavac) Orcid 0000−0002−3060−7052 (F. Engel); 0000-0003-4647-7886 (M. Vanin); 0000−0002−7881−3285 (N. Krdzavac)
From the above list of available data sources, we have decided to use in the TiVA and the BACI in this research. This is because they have a large overlap in trade flows. The BACI publishes data on bilateral trade flows at the product level [7]. The BACI database contains information on product names and the corresponding Harmonized System (HS) nomenclature for trade, export and import country codes, trade volume, trade value, unit of measure, value and annual data. With their intersection and divergent information, the TiVA and BACI databases, when integrated, complement each other to form an information-rich basis for analyzing the resilience of many different supply chains.
However, both the TiVA and BACI databases have non-binary relationships between entities. For example, the domestic value added content of gross export (code name exgr_dva) indicator has trade, which consists of trade amount, trade value and product name. This means that the exgr_dva has values for different aspects of the existing trade relation. Its trade and amount values are decimal numbers, and product names are strings. Such non-binary relations are problematic when developing models that integrate different data sources using only binary relations. We propose the use of the Web Ontology Language (OWL) [8] to model the four and three dimensional indicators, and thus provide a framework for semantic data representation and querying. In this work, we address the following research questions (RQ):
RQ1: Can we apply n-ary relations [9] to overcome the challenge of developing a model that integrates existing data sources related to supply chains? RQ2: How can federated querying be leveraged to efficiently retrieve information from the integrated ontology model concerning global supply chains? RQ3: How do we ensure interoperability between different industry classification standards used in these data sources?
To address the outlined research question, the contributions of this paper are:
• To document the value and trade flows (VTF) ontology (RQ1), the implementation of mappings between the TiVA and the International Standard Industrial Classification of all Economic Activities (ISIC) Rev. 4 industry code classification schemes (RQ3), and the VTF Knowledge Graph (VTF KG) (RQ1). • To outline the implementation details of REST API for querying the TiVA Knowledge Graph (TiVA KG) (RQ2). • To describe the implementation of federated query against the TiVA KG and BACI Knowledge Graph (BACI KG) to enrich value and trade flows data with data about products (RQ2).
The paper is structured as follows. An overview of the TiVA indicators with four and three dimensions is provided in section 2. Related work on monitoring supply chain resilience is discussed in Section 3 following by Section 4, which describes the VTF KG. Within this section, the sub section 4.1 describes the VTF ontology. Section 5 describes the implementation of the REST API for querying the TiVA KG, and the federated query against the TiVA KG and the BACI KG to enrich value and trade flows results with information about products. The Section 6 presents conclusion and future work. The Appendix section lists all the concept and role inclusion axioms of the VTF knowledge base (KB) and federated SPARQL query.
This section provides a brief description of four and three dimensional TiVA indicators [10]. The Economic Cooperation and Development (OECD) published a guide to TiVA [10,11] that outlines how to measure trends in global value chains. Table 1 provides a summary of selected indicators with four and three dimensions used in this paper. The indicator code name is given at the top of each column. Each TIVA dimension consists of a country code (C) or an industry code (I). The tuples (C, I), (I), and (C) denote the country code or the industry code or both that belong to one of the TIVA perspectives. For each indicator, the number of country and industry codes should be equal to the number of dimensions. For example, the four dimensional indicator fdva_bsci (see Table 1) has two country code values and two industry code values. In this paper, all indicators have their value expressed in the USD currency and the year is set to 2018. The code name fdva_bsci refers to the origin of value added in final demand, which is a four-dimensional indicator. It shows how the value of final demand and services consumed within a country is derived from the accumulation of values produced by several industries in different countries [10]. The value added origin and the final demand are determined by the country and industry codes. The origin of value added in the gross exports dataset, identified by the code name exgr_bsci, provides estimates of the total gross exports grouped by each exporting industry in a country. The estimates are broken down by the value added generated by the originating industry and country [10]. Both value added origin and gross exports are determined by the respective country and industry codes. The origin of the value added in gross imports with the code name imgr_bsci links the country's imports with the country of origin of the exports of goods and services of the exporting country [10]. In this indicator, the value added origin and imports are determined by country code, but exports are determined by the country and industry codes. The gross exports by origin of the value added and final destination, denoted by the code name fd_exgr_va, shows the value added from the source country that is embodied in the exports of an exporting country that ends up in the final destination country [10]. In this indicator, the value added origin and final demand are described by the source country code, and exports are identified by the country and industry codes. The domestic value added content of gross exports is a three dimensional indicator with the code exgr_dva (see page 19 in [10]). This indicator is described by country and industry codes in the export dimension and the country code in the import dimension. It means that the industry of an exporting country to the partner country in the import dimension represents the exported value added generated in the economy of the exporting country. This indicator excludes intra regional trade and intra regional value added flows [10]. All indicators of global flows of goods and services can be linked in more than two hundred billion combinations as described on page 15 of the TiVA guide [10]. For example, the origin of value added is the Chilean copper industry. German exports of auto parts embodied Chilean copper. The Chinese automotive industry imports German auto parts. Finally, the European Union has a final demand for cars assembled in China.
Several studies have used knowledge graphs and visualization approaches to monitor supply chain resilience. A recently published market convergence prediction framework [12] uses the chain knowledge graph to improve supply chain management through network resilience experiments. The knowledge graph facilitates cross-domain information connectivity for better decision making. The framework visualizes the interconnections and collaborative relationships between companies in each industry [12]. A knowledge graph-based risk management framework (SCRM) [13] for supply chain resilience is developed. The framework includes a knowledge graph for monitoring risks and long-term disruptions. The constructed knowledge graph contains 2.5 million entities. The framework applies knowledge retrieval, data visualization analysis, risk monitoring, and early warning to supply chain risk management. Many knowledge graphs suffer from incompleteness, which affects link prediction. To predict missing information and identify critical entities in the supply network, the knowledge graph completion methods are applied to link prediction [14].
The main purpose of this section is to describe the VTF KG and to address research questions RQ1 and RQ3. The VTF KG is federation of TiVA KG and the BACI KG. The VTF KG comprises of:
• The VTF ontology that is available at https://schema.coypu.org/vtf/1.4.
• The TiVA and the ISIC Rev. 4 industry code thesauruses, and the mappings between them.
• Individual assertions of VTF ontology derived from TiVA CSV files. The TiVA KG SPARQL endpoint is available at https://tiva.coypu.org/tiva. • The BACI KG created by CoyPu partners.
This section describes how to encode the TiVA [10] To elegantly express the TiVA indicators with four and three dimensions in the VTF KB, we use a Description Logic (DL) syntax [15]. The Appendix section specifies the Concept Inclusion (CI) and Role Inclusion (RI) axioms, including domain and range restrictions on role names in the VTF KB. The tеrm trade in value added refers to a set of indicators used to understand global production networks and supply chains [10]. These indicators are divided into several groups according to the number of dimensions. The VTF KB implements GrossExports, OriginOfValueAdded-FinalDemand, OriginOfValueAddedGrossExports, OriginOfValueAddedGrossImports, Domestic-ValueAddedContentOfGrossExports as concept names. These concept names are subsumed by the TradeInValueAdded concept name, as expressed in the CI from 8 to 13 in the Appendix section. These concept names are expressed as the range side of the corresponding role names, as expressed in the CI 23 through 25, and 31 in the Appendix section. The term industry sector classification scheme refers to a systematic approach to assigning classifiers to organizations based on their industry sector codes. To express this term in the VTF KB, we use the concept name IndustrySectorClassificationScheme from the Financial Industry Business Ontology (FIBO) [16]. The term ISIC Rev.4 is expressed in the VTF KB as the ISIC4 concept name and is subsumed by the IndustrySectorClassificationSchema concept name (see CI 21 and 22 in the Appendix section). The VTF ontology is derived from the VTF KB and it is implemented using the OWL2 language [17] to support integration and semantic querying of different data sources. Concept names from the VTF KB are implemented as classes in the VTF ontology and role names are implemented as object properties. The VTF ontology currently contains 16 classes, 15 object properties, and 23 logical axioms. However, datatype properties have not yet been implemented. The ontology is being reused to implement the domestic value added content of gross exports indicator (see exgr_dva code in Table 1), within the COY ontology [18]. There is a n-ary relation between trade and domestic value added content of gross exports terms that is problematic to represent in ontology using binary relations because there are quantitative values describing this relation that are type of decimal number or string [9]. To overcome this challenge, the n-ary relation ontology design pattern is used [9,19] by creating the concept name Trade and the role name hasTrade (see CI 3, 19, 20 and 28, RI 36 and 37 in the Appendix section). The concept names TradeAmount (CI 7 in the Appendix section), Product (CI 2 in the Appendix section), TradeValue (CI 5 in the Appendix section) and the role names hasTradeAmount, hasTradeProduct, hasTradeValue, including range restrictions for these role names are created to express quantitative values of Trade (see CI 15,16,17,18,30,33,34 in Appendix section). The Trade concept name is specified using reflexivity restriction on isTrade role name (see CIs 19 and 20 in the Appendix section). To link DomesticValueAddedContentOfGrossExport, TradeAmount, Product and TradeValue concept names, the RI 36, 37, 38, 39, and 40 in the Appendix section are created by using isTrade , hasTradeAmount, hasTradeProduct, hasTradeValue role names. The evaluation of the VTF ontology includes tests for accuracy, completeness, computational efficiency, consistency, and coherence [20]. The accuracy test has been passed, and there are no illegal re-declarations of entities within the VTF ontology. However, the completeness test has not been passed, because the full list of value added origin indicators [10] has not been implemented. The VTF ontology implements four indicators with four dimensions among more than 40 of the indicators listed in Table 3.1 in [10]. The computational efficiency test shows that the DL expressivity of the VTF ontology is equivalent to 𝐴𝐿𝑅𝐼 DLs, which is between DL-Lite [21] and 𝑆𝑅𝑂𝐼 𝑄 DLs [22]. This means that the HermiT [8] reasoner is able to classify the VTF ontology. The reasoner detects that the ontology is consistent and coherent.
This section describes the implementation of thesauruses for the TiVA and the ISIC Rev. 4 industry sector codes using an ontology-based approach, including the implementation of mappings between them. This subsection addresses research question RQ3. The Simple Knowledge Organization System (SKOS) [23] is used to serialize mapping between the thesauruses. The result of this implementation is:
• The TiVA industry sector codes thesaurus.
• The ISIC Rev. 4 industry sector codes thesaurus.
• Automatically produced mapping between these two thesauruses.
The CoyPu GitLab repository2 provides the complete implementation and an explanation of how to reproduce the listed thesauruses and mappings. We start by using Table A.3, which was published in [10], and the ISIC Rev. 4 document, which is available in [24]. Table A.3 provides a comprehensive list of all TiVA industry sector codes and their correspondence to ISIC Rev. 4 industry sector codes. The rationale for the approach used in this section is twofold. It can be used to produce and validate mappings between any other industry codes that are not explicitly given, as shown in Table A. 3. The other reason is that the ISIC Rev. 4 industry sector codes may change over time and this generic solution can be used to produce and validate mappings between the TiVA industry sector codes and the ISIC Rev. 4 industry sector code based on these changes. Figure 2 shows the implementation workflow. The first task is to automatically generate SKOS-based thesaurus from ISIC Rev. 4 industry sector codes 2 . This is achieved by mapping from the ISTC Rev. 4 CSV file to the corresponding TTL file using RDFizer [25]. The resulting SKOS-based thesaurus is validated against SKOS shapes by using the TopBraid SHACL [26] engine. In the next step, marked with number 2 in Figure 2, the TiVA SKOS-based thesaurus is automatically created and validated against the SKOS shapes. Table A.3 published in [10] shows the correspondence between the TiVA industry sector codes and the ISIC Rev. 4 industry sector codes, but does not show the semantic relations between these two sets of industry codes. In this work, these semantic relations are automatically generated by the LogMap [27] matching tool. The tool accepts the ISIC Rev. 4 and TiVA as source and target thesauruses respectively. As a result, the tool generates 38 mappings between the source and target thesauruses. The LogMap tool did not produce conflictive mappings between these two thesauruses. Each mapping contains a SKOS concept from the ISIC Rev. 4 thesaurus, a SKOS concept from the TiVA thesaurus, the type of mapping between these two SKOS concepts, the mapping direction, and the mapping confidence. This information is stored in a CSV file available in the CoyPu Gitlab repository 2 . In the 4th and 5th steps shown in Figure 2, the RDFizer tool converts the CSV file containing information about mappings into a TTL file. The resulting TTL file is validated against the SKOS shapes using the TopBraid SHACL engine.
The TiVA KG is implemented by processing the raw data, which are available as CSV files [6] and representing four dimensional indicators listed in Table 1. The raw data consists of industry and country code names, which are string types, and value and year, which are numbers. The Python code transforms raw data into TTL files consisting of individual assertions, which are instances of the VTF ontology schema. Table 2 summarizes the size of the ontology files generated for each TiVA indicator raw file. The final step is to load generated TTL files, VTF ontology schema, TiVA and ISIC Rev. 4 industry sector code thesauruses, and the produced and validated mappings between these two thesauruses [28] into a remote triple store. The size of the VTF KG is 257GB and it contains 1128749054 triples.
Data on the domestic value added content of gross exports (exgr_dva code name) three dimensional indicator are stored in the BACI KG 3 . The BACI KG is populated from the international trade at product level database [7]. It contains information on exporting and importing country codes, product names, product codes, trade amount value, trade amount, year of trade. The BACI KG contains 185251116 triples and was created by CoyPu project partners.
In this section, we discuss the design and implementation of the REST API to query VTF KG. It addresses research question RQ2. The pipeline consists of two main building blocks, namely Backend, Data Storage , which are shown in Figure 3.
The output of the Backend module is a JSON file as a result of querying VTF KG by executing parameterized SPARQL queries against a remote TiVA or BACI SPARQL endpoints [28]. The Swagger interface 4 allows users to interact with the API implemented in the Backend module by passing parameters for each operation. Users can also implement their own client-side solutions using this RESTful API. Figure 4 shows an example of a parameterized SPARQL query. The parameter in this query is a trade location, highlighted in red, of the value added origin in the origin of value added in gross imports indicator. The result of this query is a JSON file consisting of exports trade location, exports industry code, imports trade location, value and year for origin of value added in gross imports.
The aim of the previous sections was to present our work with regard to the integration of data sources. This section addresses research question RQ2, essentially how we can utilise integrated data sources. As a proof of example, we have created a federated SPARQL query, available in Appendix section, to get information about product trade information, about location of exporting and importing country, product code and name and value (in the USD currency).
The implementation of the federated query involves a SPARQL query executed programmatically against the two SPARQL endpoints of TIVA KG and BACI KG (see Data Storage component shown in Figure 3). A JSON object as a result of the execution of this federated query contains the value, year, exporting and importing trade location of the imgr_bsci indicator available in TiVA KG (see 27 and 28 rows in the federated SPARQL query). These exporting and importing trade locations must match the importing and exporting trade locations of the exgr_dva indicator available in BACI KG (see rows 41 and 42 in the federated SPARQL query). Based on this match, the resulting JSON object also contains the product name, product code, quantity value and year of trade available in BACI KG.
In this work all research questions are addressed. This paper shows valuable results that are the basis for deeper analysis of international trade flows via building more federated SPARQL queries and implementation of a dashboard that should dynamically generate charts using implemented REST API. The OECD forum5 recently discussed four key issues 6 for resilient supply chains. To address one of these issues, this paper presents VTF KG and a RESTful API. The OECD discusses the need to implement policies that strengthen the resilience of supply chains. One of the key policy actions is to determine government role, which includes the international exchange of information7 . This paper addresses this issue by enabling services and tools to share information about trade flows using a RESTful API and to perform federated queries against TiVA and BACI knowledge graphs. We observe gaps in this work that should be addressed in further development such as to infer missing information in the VTF KG and to solve the incompleteness of the VTF ontology.
1 PREFIX r d f : < h t t p : / / www. w3 . o r g / 1 9 9 9 / 0 2 / 2 2 − r d f − s y n t a x − ns #> 2 PREFIX v t f : < h t t p s : / / schema . coypu . o r g / v t f #> 3 PREFIX r d f s : < h t t p : / / www. w3 . o r g / 2 0 0 0 / 0 1 / r d f − schema #> 12 ? e x g r d v a v t f : h a s I m p o r t ? i m p o r t .
13 ? e x g r d v a v t f : h a s E x p o r t ? e x p o r t .
14 ? e x g r d v a v t f : h a s T r a d e ? t r a d e .
The research has received funding from the Federal Ministry for Economic Affairs and Energy of Germany in the project Cognitive Economy Intelligence Plattform für die Resilienz wirtschaftlicher Ökosysteme -CoyPu (project number 01MK21007[A-L]).
Value and trade flows (VTF) knowledge base (KB) expressed in Description Logic (DL) syntax: