Informatica

Supply chains are inter-organizational networks among different companies, including suppliers, manufacturers, distributors, and customers, operating through a value chain in the industry (Capaldo and Giannoccaro, 2015; Surana et al., 2005). Working together, both internally within an organization and externally with different companies – aka supply chain partners, – is essential for effective Supply Chain Management (SCM) (Emmett and Crocker, 2016). SCM integrates key business processes from end-users through original providers who supply customers with products, services, and information to add value at the lowest cost to the supply chain (Christopher, 2005; Cooper et al., 1997).

The structure of efficient supply chain systems could be captured by complex network models (Surana et al., 2005; Hearnshaw and Wilson, 2013; Pathak et al., 2007), categorized into four main classes: social networks, information networks, technological networks, and biological networks (Newman, 2003). A Supply Chain Network (SCN) fits the definition of an information network, which is formed by interacting components. In most studies on network science, each node has the same type of objects (e.g. companies), and each link has the same type of connections (e.g. material flow). Such networks are called homogeneous information networks. On the other hand, most real-world networks follow the definition of Heterogeneous Information Networks (HIN), where nodes and links have different types (Sun and Han, 2012). Since the existing studies for the analysis of homogeneous information networks in the literature cannot be applied directly to HINs, the analysis of HINs has emerged as an area of research gaining importance (Shi et al., 2017). For example, in a SCN, while nodes represent customers, suppliers, products, etc., links represent relations like supply/supplied-by, purchase/purchased-by. Examining all nodes by assuming they are similar could result in some critical information loss. It is thus desirable to model SCNs as HINs, with a structure where organizations and their relationships have different types, like most real-world networks. By defining components such as raw materials and finished products as other types of nodes, SCNs become capable of presenting the indirect interactions between an organization’s suppliers and customers. Furthermore, the relevant feature vectors for both nodes and links will enrich the information content of the network.

Four tasks have been highlighted in HIN literature: (1) classification (Shehnepoor et al., 2017; Zhang et al., 2021), which predicts the label of all types of objects based on some objects with known labels, (2) clustering (Sun et al., 2012; Zhou et al., 2020), which groups objects so that similar ones are in the same cluster while dissimilar ones are in different clusters, (3) link prediction (Davis et al., 2013; Wang et al., 2021), which estimates the likelihood of the existence of a link between two nodes, and (4) similarity (Shi et al., 2014; Wang et al., 2016a), which evaluates the similarity of objects. Beyond these foundational tasks, advanced tasks include (1) recommendation (Shi et al., 2019; Xing et al., 2025), which utilize intricate network relationships, employing meta-path and meta-graph frameworks to enhance individualized recommendation, (2) representation learning (Dong et al., 2017; Zhang et al., 2025), which aims to encapsulate structural and semantic information into low-dimensional vectors to facilitate various downstream tasks, and (3) information diffusion (Zhang et al., 2016), which examines the propagation of information, behaviours, or influence across diverse networks, yielding essential insights into network dynamics and forecasting. Foundational tasks create the essential analytical framework for comprehending HINs, whereas advanced tasks utilize deeper structural and semantic insights, enabling complex analytical skills and improved prediction performance.

As for SCM, statistical analysis, simulation, and optimization are the headings highlighted in the literature (Narayanan et al., 2015; Tiwari et al., 2018). There are two types of statistical analysis techniques: descriptive and predictive. The former employs data to extract information about the characteristics of the SCN. The latter makes future predictions based on historical data. The simulation enables developers to determine how the system is affected under various system configurations and levels of complexity. The optimization derives knowledge from a complex system with numerous factors and constraints. Although it is beneficial to solve SCM problems under the three categories, the studies on SCNs are rare and related to SCNs being considered complex adaptive networks, not as HINs (Surana et al., 2005; Choi et al., 2001; Li et al., 2017).

Linguistic Summarization (LS), proposed by Yager (1982), can be described as a descriptive data summarization method. Understanding and interpreting the results of analytical methods is challenging for people who do not have sufficient knowledge about the subject. By LS, the obtained information is expressed with natural language that allows better understanding. By way of illustration, “Most of the trusted suppliers supply materials in low lead times. $[0.78]$” is a LS with an assigned truth degree which takes value in the unit interval. If the generated summary form is valid, the truth degree is closer to one. LS has become a popular approach since it may bring valuable information from data (Ramos-Soto et al., 2016; Conde-Clemente et al., 2017; Kaczmarek-Majer and Hryniewicz, 2019; Nguyen et al., 2021; Özdoğan et al., 2021). LS is a multidimensional framework that integrates fuzzy quantifiers, linguistic variables, and aggregation functions to generate understandable data summaries. Although many works have been done in the field of the supply chain, quite limited studies generate LSs from network data. Aydoğan et al. (2021) used LS to support strategic decisions in single objective and bi-objective supply network design, in which the stages of the supply chain show the nodes, and the material flows between them show the relations.

The performance of an SCN can be measured in terms of resource, output, and flexibility (Beamon, 1999). The resources may consider the costs of manufacturing, distribution, inventory, and return on investment for high-level efficiency. The output may consider sales, profit, fill rate, on-time delivery, stock out cases, response time to customer, lead time, shipping errors, and customer complaints about the degree of customer service. The flexibility may take the ability to respond to a changing environment. As partnership quality in SCNs indicates Supply Chain Network Performance (SCNP), it has received significant attention in the literature (Yang et al., 2020; Zhou et al., 2016; Li et al., 2015). However, most previous studies on relationship dynamics are qualitative analyses based on querying hypotheses. Though it will be examined in detail in the literature review section, we would like to exemplify the subject here. For example, Rodriguez-Lopez et al. (2017) point out the hypothesis that “The higher the trust, the better the economic performance”. While the hypothesis is compatible in some studies, like in Akhtar and Khan (2015), conflicting with others, like in Arora et al. (2021). Therefore, quantitative ways are needed to reveal the interaction between supply network relations and SCNP. Given this context, the study’s main hypothesis is that trust between supply chain network participants significantly and favourably affects the network’s performance, especially when it comes to organizational and operational results. By this means, the validity of qualitative studies can be tested with quantitative studies, like truth degree in LS.

Despite the increasing interest in trust dynamics within SCNs, previous research has primarily depended on qualitative assessments or uniform modelling methods that neglect the structural complexity of actual systems. This study’s originality is in the amalgamation of linguistic summarization and HIN modelling, a combination not previously utilized in supply chain trust research. This connection facilitates the extraction of interpretable, semantically rich summaries from intricate, multi-type relational data, so connecting data-driven analysis with managerial insight. Additionally, by substantiating literature-derived assumptions using the truth values of language summaries, the study offers a quantitative yet comprehensible alternative to conventional trust-performance modelling.

This paper aims to generate and evaluate LSs on how relationships affect SCNP. The supply network is modelled as an HIN to minimize information loss. Generated summaries are used to test the validity of the qualitative studies in the existing literature. There are several essential areas to which this study makes an original contribution:

The remainder of the study is organized as follows: Section 2 presents the extant studies that reveal the relationships affecting performance in SCNs. Our LS procedure of HINs is explained in detail in Section 3. The methodology overview is summarized in a diagram in Section 4. A realistic case study is conducted, and LS results are compared with existing literature in Section 5. The results are discussed in Section 6. The paper is concluded with Section 7.

Research into SCM consists of statistical analysis, simulation, and optimization (Tiwari et al., 2018; Wang et al., 2016b). Many statistical analysis studies have been conducted to date to investigate how the trust relationship among SCN members affects SCNP. These studies established hypotheses about the effect of trust on SCNP, such as “Inter-firm trust positively affects a firm’s operational performance (OPP)” (Shi and Liao, 2015). Similar hypotheses were tested using questionnaires applied to companies. The research question that started the systematic review was, “How does trust affect the performance of supply chain networks?” To do a systematic literature review, this study is advanced by creating a Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) list (Liberati et al., 2009). Due to its constant coverage of high-quality, peer-reviewed journal articles in engineering and management sciences, the Web of Science Core Collection was chosen as the only database. The last thirteen year of the literature (January 2012–May 2025) is considered to focus on up-to-date studies, and the review process is summarized as follows:

The evaluation only included papers that specifically looked at the connection between SCNP and trust; studies that didn’t specifically address this relationship were not included. Two authors separately reviewed the titles and abstracts of the publications that were retrieved in order to minimize subjectivity in the selection process. Any discrepancies were then discussed and settled until an agreement was achieved. Based on the PRISMA list, the final 32 articles are summarized in Table 1 regarding a trust-SCNP relationship.

Table 1 provides a systematic summary of chosen studies examining the impact of trust on SCNP. The table classifies the studies according to the type of effect examined (direct, indirect, or moderating), the nature of the trust variable (e.g. inter-firm, supplier, or customer trust), the dimension of performance assessed (operational or organizational), and the specific performance metrics utilized (such as lead time, profitability, delivery reliability, or innovation). The concluding column encapsulates the study’s findings regarding the impact of trust on performance, indicating whether the effect is favourable, negative, or mixed. This classification elucidates the operationalization and measurement of trust in the literature, emphasizing both consistent trends and contradictory results. The table facilitates the identification of theoretical and methodological shortcomings addressed in the present work.

Based on the examined literature, numerous significant research gaps may be discerned: (i) although most studies acknowledge the significance of trust in supply chain performance, the operationalization of trust exhibits considerable variability and is frequently constrained to unidirectional or static perspectives, (ii) the majority of analyses depend on perceived relationships obtained from surveys, rather than on actual or structural network data, (iii) there is little emphasis on capturing the heterogeneous and multi-relational characteristics of supply chain interactions; current models frequently depict the supply chain as a homogenous or linear system, and (iv) the interpretability of findings is seldom considered; few studies provide comprehensible or linguistically articulate insights to assist decision-makers. The deficiencies shown in the summary findings of Table 1 and the extensive literature discourse necessitate the strategy advocated in this study.

This section presents the fundamental elements of the suggested methodology. The text initially outlines the conventional LS methodology, subsequently introducing an enhanced LS framework tailored for HINs. Ultimately, it addresses the interpretability facet of LS, highlighting the semantic clarity and decision-making assistance that LSs can provide in intricate relational contexts like supply chains. The theoretical backdrop from the literature is compiled in Section 3.1, our suggested adaption to heterogeneous information networks is presented in Section 3.2, and a previously suggested but untested method is implemented in Section 3.3, which makes up this study’s original contribution.

This section delineates the methodology framework employed to assess the impact of trust on SCNP, as illustrated in Fig. 2. The procedure involves a sequence of organized phases underpinned by a blend of diverse network modelling, fuzzy clustering, and LS. The suggested method facilitates the creation of interpretable, quantified assertions based on network-structured data.

Box 1 presents the input data as an HIN consisting of several node types (e.g. suppliers, products) and link types (e.g. supply, purchase). In addition to being a different type, every node and link may possess numerical or categorical properties.

In Box 2, numerical variables like profitability or cost are segmented into fuzzy clusters utilizing the Fuzzy C-Means (FCM) algorithm (Bezdek et al., 1984). FCM was used to produce linguistic membership degrees through a data-driven approach. This enables fuzzy sets (e.g. “low”, “medium”, “high”) to accurately represent the data distribution, hence diminishing reliance on subjective, expert-defined thresholds. The resultant membership functions are therefore both comprehensible and empirically substantiated. Membership degrees to linguistic categories such as low, medium, and high are computed and recorded as fuzzy sets (e.g. ${\mu _{\text{high}}}({x_{i}})$), which then function as summarizers in the LS process.

In Box 3, metapaths that serve as the structural equivalent of LS statements are extracted. A metapath delineates a schema-level trajectory linking nodes via significant relationships, exemplified by Supplier–Supplies–Raw material. These pathways encapsulate semantically interpretable sequences inside the network. For instance, the metapath Supplier → Supplies → Raw material may substantiate a statement such as: “Most of the highly trusted suppliers supply sheet-metal raw materials”.

In Box 4, sentence structures involving two- and three-node metapaths are generated. Each sentence structure consists of a quantifier Q (e.g. most), a summarizer or qualifier A or B (e.g. high profitability), and a truth degree ($\textit{TD}$ $\in [0,1]$).

In Box 5, these sentence structures are evaluated using iteration-based methods. The truth degree $\textit{TD}$ is calculated for each summary, based on fuzzy set operations and network semantics.

In Box 6, all possible combinations are generated and filtered using a predefined threshold τ (e.g. $\textit{TD}\lt \tau $). Sentences with low truth degrees are discarded. The remaining summaries are processed for interpretability (e.g. consistency, similarity).

Finally, in Box 7, the validated LSs are compared with theoretical hypotheses from the literature. Each summary is then discussed as either supporting or contradicting existing claims, enabling a transparent and data-driven validation framework.

The chosen methodology was picked for its capacity to reconcile semantic interpretability with structural modelling efficacy. LS offers clear, comprehensible summaries for decision-makers, whereas FCM guarantees that linguistic concepts correspond to inherent groupings within the data. The HIN model facilitates multi-relational supply chain representations, allowing for phrase structures that incorporate trust-based pathways and performance-associated nodes. This combination was selected over alternative black-box or rule-based methodologies because it aligns with the data’s complexity and the study’s interpretability objectives.

The novelty of this methodology lies in the integration of fuzzy LS with HINs—an approach that enables interpretable reasoning over complex relational data, which is rarely achieved in existing supply chain analytics literature.

SCNs are highly needed for analysis as they contain rich information from many stages between the raw material supplier and the customer who purchased the finished product. Most real-world problems are attempted to be solved in theory with several assumptions, detracting solutions from reality. Apart from assuming that nodes and relations are of the same kind, modelling SCNs as HINs is a technique that should be used since minimizing information losses. This section of the study offers modelling of rich and complicated supply network systems and the discovery of meaningful summaries suitable for human perception.

This study presents an innovative amalgamation of LS and HIN modelling to examine the impact of trust on SCNP. The findings indicate that consumer trust significantly influences ORP metrics, including profitability and volume, although its impact on OPP is little. These findings largely validate prior literature indicating a favourable association between trust and performance, while also revealing discrepancies that may stem from contextual variables or data-driven analysis.

This study provides an objective and measurable approach to hypothesis verification by calculating the truth degrees of language summaries, in contrast to typical studies reliant on surveys and perceived associations. This data-driven yet comprehensible method connects intricate analytics with management insight, fostering transparency and trust in decision-making.

Furthermore, the suggested method facilitates digital transformation objectives by enabling firms to uncover concealed patterns in supply chain data through a comprehensible manner. This is especially crucial for non-technical users, allowing them to participate in network-level assessments and strategic development.

A principal advantage of the technique is its capacity to manage heterogeneous, multi-type relational data without reducing information to simplified models. Nevertheless, the results must be understood in light of the study’s limitations, which include its dependence on a singular real-world dataset from the automotive industry and the subjective nature of trust evaluations. Future research may extend this framework to various industries, implement dynamic trust modelling, or integrate real-time data.

The method improves the interpretability, traceability, and utility of supply chain analytics within the framework of digital transformation.

Several significant discoveries resulted from the analysis:

The insights of this study apply to both academic and industrial sectors. Industrial supply chain managers and analysts can use the proposed methodology to assess how network trust affects organizational and operational performance using enterprise-level data (e.g. ERP systems). In complicated supply chains including automotive, home appliances, and electrical equipment, reliable coordination is crucial. A reproducible, data-driven strategy that can be applied to multiple businesses and circumstances helps academics investigate trust and performance. Thus, the findings inform management decision-making and promote supply chain trust concept.

Notwithstanding the merits of the suggested methodology, many limitations must be recognized. The study relies on a singular real-world dataset from the automobile sector, thereby constraining the generalizability of the results. Secondly, trust levels were deduced rather than actually measured, which may impact their accuracy. The linguistic labels and fuzzy clusters employed in the summarization process depend on parameter configurations that may involve a degree of subjectivity. The model is presently static and fails to account for the temporal dynamics of trust relationships across time. These constraints highlight significant prospects for future endeavours, encompassing the incorporation of dynamic trust modelling, real-time data streams, and adaptive fuzzy systems across diverse fields.

This paper set out to determine the effects of trust on SCNP with LS over HIN. In this study, SCN was modelled as an HIN for the first time. The lack of information caused by excluding object types such as products and raw materials in SCN has been eliminated. With the advantage of HIN, the relationships were revealed from the interactions between node attributes (i.e. trust level) and link attributes (i.e. lead time). A systematic literature review was done, and hypotheses for the direct, indirect, and moderating effect of trust on OPP and ORP were extracted. An application was performed with a quantitative dataset in the automotive industry to eliminate the shortcomings of existing studies based on personal perceptions. While the summary forms containing nested quantified nodes were expressed by iteration-based polyadic quantification in line with linguistic logic, they were evaluated by semi-fuzzy quantifier-based method. The study concluded that consumer trust had a greater influence on ORP than OPP. Furthermore, the proposed methodology effectively tested or challenged existing ideas in a data-driven and interpretable fashion. The findings indicate that trust modelling in supply chains is enhanced by both structural representation and linguistic interpretation, presenting a promising avenue for future research and managerial applications. Given the growing focus on digital transformation within supply chains, the suggested method enhances this framework by providing a clear and interpretable approach for examining trust-related dynamics. This facilitates more intelligent, data-driven decisions while maintaining clarity, which is essential in multi-stakeholder contexts.

The results show that customer trust effectively affects ORP for profitability and volume measure. The research has also shown that neither customer nor supplier trust is effective on OPP. The results indicate that trust throughout the supply chain is consistently correlated with enhanced SCNP, especially regarding profitability indicators at organizational level. This validates literature-derived hypotheses and illustrates the utility of LS for evaluating performance-related trends in diverse networks. Although findings are the results obtained by evaluating with a quantitative method, they are limited because they belong only to one company operating in the automotive sector. Furthermore, the results are encouraging, it is crucial to acknowledge that the present solution is constrained in scope and presumes static, inferred trust connections. Subsequent research may improve this approach by integrating real-time trust data, examining further industries, and optimizing the linguistic parameterization process. It should be noted that different outcomes could be obtained when the research is repeated with additional variables for other sectors. Further research might be carried out in another sector.