Driving Supply Chain Transformation with IoT and AI Integration: A Dual Approach Using Bibliometric Analysis and Topic Modeling

Abstract

The objective of this study is to conduct an analysis of the scientific literature on the application of the Internet of Things (IoT) and artificial intelligence (AI) in enhancing supply chain operations. This research applies a dual approach combining bibliometric analysis and topic modeling to explore both quantitative citation trends and qualitative thematic insights. By examining 810 qualified articles, published between 2011 and 2024, this research aims to identify the main topics, key authors, influential sources, and the most-cited articles within the literature. The study addresses critical research questions on the state of IoT and AI integration into supply chains and the role of these technologies in resolving digital supply chain management challenges. The convergence of IoT and AI holds immense potential to redefine supply chain management practices, improving productivity, visibility, and sustainability in interconnected global supply chains. This research not only highlights the continuous evolution of the supply chain field in light of Industry 4.0 technologies—such as machine learning, big data analytics, cloud computing, cyber–physical systems, and 5G networks—but also provides an updated overview of advanced IoT and AI technologies currently applied in supply chain operations, documenting their evolution from rudimentary stages to their current state of advancement.

1. Introduction

The new technologies associated with Industry 4.0, such as AI, blockchains, big data analysis, and the IoT, have the potential to significantly reshape the future of business and supply chain management [1]. The implementation of IoT in supply chain management (SCM) addresses various issues linked to traditional SCM processes, such as overstocking, understocking, delivery delays, and real-time information dissemination challenges [2]. IoT refers to an information network that integrates sensors embedded in or attached to a wide range of physical objects, including consumer goods, pallets of goods, everyday tools, household appliances, and industrial machinery [3]. Some of the most commonly used IoT devices in SCM include Radio Frequency Identification (RFID) tags and readers, GPS trackers, sensors, actuators, smart shelves, robotic pickers, wearables, and drones. Companies from various industries are increasingly adopting IoT technology to enhance different aspects of their supply chains, ranging from manufacturing and logistics to inventory management and quality control, leading to enhanced efficiency, visibility, and decision-making capabilities. For example, Amazon improved its order fulfilment operations through the implementation of an IoT-based warehouse management system [4]. This system incorporates RFID, weight sensors, ultrasonic modules, and wearables with health monitoring features for employees. These technologies are seamlessly integrated with Amazon Web Services, yielding a cost-effective and efficient solution that enhances warehouse management and optimizes the supply chain. This system not only facilitates package tracking but also automates inventory flow, optimizing order processing and delivery.

While IoT holds the promise of enhancing supply chain performance, its integration is not devoid of complexities and challenges [5]. Various corporate and technological issues have emerged around IoT infrastructure that necessitate resolution before widespread adoption [6]. Critical challenges in IoT-based supply chains include data quality and reliability, privacy and security concerns, data/information silos, and data integration [7]. Given that IoT devices continuously collect and transmit sensitive data, they are vulnerable to cyber-attacks and data breaches, potentially resulting in severe consequences such as compromised intellectual property, financial losses, and harm to a company’s reputation [5].

Several studies have recognized and discussed the positive impact that AI can have on various SCM functions by enhancing operational efficiency, performance, and sustainability [8]. Machine learning (ML), one of the most important sub-branches of AI, imitates the learning mechanism of the human brain, and its rapid development in recent years has led to the introduction of many applications [9]. The utilization of ML can assist in overcoming some of the most common challenges encountered by IoT-enabled supply chains. The application of ML algorithms in Intrusion Detection Systems (IDSs) exemplifies these solutions. IDSs, which can be devices or software applications, are employed to monitor computer systems or networks for activities that deviate from normal operations, potentially violating policies [10,11]. In the context of IoT-enabled supply chain networks, ML algorithms within IDSs are crucial for identifying potential cyber threats and facilitating timely responses that preserve system integrity [12]. Another important application of AI is observed in the processing of huge quantities of data generated by the IoT devices in IoT-enabled SCM. In agri-food supply chains, IoT devices like sensors, RFID tags, and connected devices furnish vast quantities of real-time data on crucial parameters, such as location, temperature, and humidity. By analyzing these data using ML algorithms, organizations can make informed decisions, predict disruptions, and optimize logistical operations. Therefore, the integration of IoT and AI has the potential to revolutionize SCM, leading to significant enhancements in productivity, visibility, and sustainability [13].

While interest in the application of IoT and AI to SCM is growing, a need remains for comprehensive studies that systematically review the academic literature, particularly highlighting key trends and topics in this discipline. The primary goal of this research is to conduct a systematic review of the literature on the integration of AI/ML and IoT within supply chain management, with a focus on identifying key trends and critical topics. Through the analysis of publication patterns, geographical research distribution, supply chain challenges, and AI/ML algorithm usage, this study aims to clarify the mutual influence between IoT and AI/ML. Furthermore, by examining the latest advancements in AI and IoT technologies within supply chain operations, tracing their development, and proposing future directions, this research seeks to fill existing gaps in the literature and offer detailed insights into this intersection. By identifying key contributors and drivers, the study lays a solid groundwork for future investigations, ultimately providing a holistic view of the intersection between AI/ML, IoT, and supply chain management. Given these considerations, the study will address the following primary research questions:

RQ 1. What is the current state of research in IoT-AI integration in supply chain management?

RQ 2. What IoT strengths and weaknesses affect AI/ML positively or negatively in the studies?

RQ 3. What AI/ML strengths and weaknesses affect IoT positively or negatively in the studies?

RQ 4. How can AI/ML and IoT address the challenges of digitalized supply chains?

RQ 5. Which AI/ML algorithms are used in the studies?

This paper is structured into five main sections to offer a thorough examination of the roles of IoT and AI in SCM operations. Section 2 outlines the data collection process, detailing the screening criteria and research methodology applied. Section 3 interprets the results derived from bibliometric analysis and topic modeling that helped to answer the research questions, RQ 1 (Section 3.1.1), RQ 2 (Section 3.1.5), RQ 3 (Section 3.1.6), RQ 4 (Section 3.2), and RQ 5 (Section 3.3). Section 4 discusses the findings of this study, offering key insights. Finally, Section 5 concludes the paper, addresses the limitations of the study, and highlights the trends for future studies.

2. Research Method

This section provides an overview of the data collection process, followed by a discussion of the techniques used for data analyses. For the purposes of this study, we used the Web of Science database, which offers comprehensive coverage of high-quality scholarly research, to collect our dataset. After obtaining the dataset, we conducted various types of bibliometric analyses, including descriptive, trend topic, and network analyses, to explore how IoT and AI technologies enhance supply chain operations. Additionally, we conducted topic modeling on the dataset to derive valuable insights that are not easily observed through traditional analyses. This combined approach provided a comprehensive perspective on the research field by combining quantitative and qualitative insights. While topic modeling adds a qualitative element by automatically detecting topics within huge datasets, bibliometric analysis provides an organized, quantitative overview of the development of a body of literature. This method minimizes researcher bias, ensures broad coverage of the field, and aids in recognizing influential works and emerging trends. These analyses addressed the research questions outlined in the previous section. Figure 1 provides a pictorial representation of the research method applied in this paper.

2.1. Data Collection

Clarivate’s Web of Science served as the main database for our data extraction process, as it provides a comprehensive and curated list of high-quality peer-reviewed articles [14]. To retrieve the articles relevant to our research topic, we applied the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework, the standard format for reporting systematic reviews. The inclusion of the PRISMA framework helped to ensure that our study would be valuable to readers by providing information on why the review was conducted, how it was performed, and what the findings were. The PRISMA flowchart in Figure 2 illustrates the three main stages of the data collection process: identification, screening, and inclusion, which are explained in detail in the following paragraphs.

The identification stage began with the development of a search query (listed in Figure 2) focused on papers exploring the application of AI and IoT in the supply chain field, ensuring the inclusion of key terms related to technologies, processes, and their applications within this domain. On 25 June 2024, we conducted our most recent search by only including articles, proceeding papers, and early-access materials written in the English language. We identified a total of 1497 publications, which included articles with the keywords from our search string in their abstracts, titles, or topics. Although the data selection process did not impose any year-based restrictions, the dataset ultimately included publications spanning from 2011 to 2024.

After identifying the dataset, we proceeded with a two-step screening process. Web of Science provides filtering criteria, including meso- and micro-level citation topics, to refine the dataset. In the first step, we excluded papers in categories such as “Geochemistry, Geophysics & Geology”, “Water Resources”, “Thermodynamics”, “Concrete Science”, “Geometrical Optics”, “Sports Science”, “Sleep Science”, “Gait & Posture”, “Cardiac Arrhythmia”, “Environmental Sciences”, “Credit Scoring”, and “Travel Behavior”. These categories were found to be irrelevant to our research topic. This refinement step resulted in a reduced dataset of 1294 documents.

The screening stage was immediately followed by the inclusion stage, ensuring that our dataset consisted solely of papers addressing the application of AI and IoT within the supply chain context. This thorough process led to our final dataset, which comprised 810 documents [15].

Thus, the data collection process yielded a final dataset of 810 documents, which served as the input for the subsequent phases of our research.

2.2. Bibliometric Analysis

Bibliometric analysis is a key tool for quantitatively assessing large volumes of literature, allowing for the identification and evaluation of the performance and evolution of a given research domain [16,17,18]. Its primary function is to assess various aspects of published research, offering insights into trends, patterns, and the impact of scholarly contributions. Biblioshiny, an open-source bibliometric tool based on RStudio (version 4.4.0), provides flexibility for tasks such as index computation, network analysis, and knowledge map creation [19,20]. Similarly, VOSviewer, another popular bibliometric tool, is widely used to visually represent key research elements, such as co-authorship, keyword analysis, and citation networks, through graphs [21].

VOSviewer, a Java-based tool, serves the purpose of generating maps rooted in network data, bibliographic data, or textual data [22]. The software enables the visualization and examination of these maps. The bibliographic data sourced from the Web of Science in plain text form were used as input for the VOSviewer software (version 1.6.20) to produce diverse bibliographic network maps, encompassing co-authorship, co-occurrence, citation, bibliographic coupling, and co-citation analyses [23].

This paper incorporated bibliometric analysis, using Biblioshiny (version 4.1.3) and VOSviewer, to provide an objective overview of the research landscape on IoT and AI in supply chain operations. Initially, this study involved analyzing publication status, such as the number and frequency of publications, and identifying key authors and institutions contributing to the field. Additionally, we aimed to track the evolution of research topics over time, identify emerging trends—such as advancements in AI for supply chain optimization—and pinpoint research gaps. We also evaluated the impact of publications through citation analysis. Section 3.1 features an in-depth discussion of the results from the bibliometric analysis of our dataset.

2.3. Topic Modeling

The dataset of 810 documents, collected in a raw and unstructured format, contained valuable insights that were not easily observable or readily analyzable. Consequently, we applied topic modeling techniques to the abstracts of our dataset’s documents, deriving insights that would have been difficult to obtain through conventional text mining methods. By employing topic modeling, we gained a probabilistic understanding of the underlying structure within these unstructured documents [24]. This approach significantly enhanced the quality and scope of analysis by uncovering latent patterns that conventional methods may have missed. The results of the topic modeling on our dataset are discussed in detail in Section 3.2.

Before applying topic modeling techniques, the dataset underwent several preprocessing steps to ensure textual data quality and consistency. First, all documents were tokenized into individual words, followed by lowercasing, to maintain uniformity. Common stop words, numbers, and punctuation were removed to eliminate irrelevant information. Next, lemmatization was applied to reduce words to their base forms, ensuring that different word inflections were treated as the same term. Afterward, a term-frequency matrix was generated, and terms with extremely low or high frequencies (common across nearly all documents) were filtered out to focus on meaningful content. This resulted in a cleaner and more manageable dataset ready for NMF algorithms to uncover underlying topics.

Topic modeling is a text mining technique within machine learning that identifies topics and their relationships within a corpus [25]. Common algorithms include Latent Semantic Analysis (LSA), Probabilistic Latent Semantic Analysis (PLSA), Latent Dirichlet Allocation (LDA), and Non-Negative Matrix Factorization (NMF) [26]. In this study, our primary focus was on two algorithms: LDA and NMF. LDA is a probability-based algorithm for topic modeling [27]. It operates as an unsupervised machine learning method that employs a parametric approach to explore term groupings and assess relationships between terms through generative probability [28]. After comparing the performance of both LDA and NMF on our dataset, we found that NMF more effectively captured distinct and localized topics, providing higher-quality insights. Also, NMF enhances topic coherence and offers practical advantages, such as reduced dependency on model assumptions and easy interpretation [29]. Therefore, NMF was chosen as the preferred algorithm for our analysis.

3. Results

This section presents the results of the bibliometric analysis conducted using Biblioshiny and VOSviewer and the NMF technique of topic modeling.

3.1. Bibliometric Analysis Using Biblioshiny and VOSviewer

The bibliometric analysis process generated comprehensive results, as detailed in the following sections. These findings encompassed annual publication numbers, the scientific contributions of nations, the most globally and locally cited publications, the most relevant keywords, and the co-occurrence and co-citation network analysis.

3.1.1. Annual Production of Publications

This section provides the answer to our first research question, RQ1, regarding the current state of research in IoT-AI integration in supply chain management. The answer to this question provides an insight into the level of popularity of this topic among researchers.

Figure 3 illustrates the annual distribution of research in the field of IoT and AI in the context of supply chain operations. The graph demonstrates a consistent upward trend, indicating a growing interest in this field and a significant increase in the number of publications over the years. Research activity in this area emerged in 2011 and remained relatively stable until 2017, with a slight increase during that period. However, since 2020, there has been a substantial surge, with the number of articles rising from approximately 67 in 2020 to 179 in 2022.

Several key studies contributed to this surge. Refs. [30,31,32] explored the integration of IoT, AI, robotics, big data analytics, and blockchains into agricultural and food supply chains, addressing both the applications and challenges of these technologies. Ref. [33] examined the implications of smart cities on supply chains, while [34] provided a review of the disruptions caused by the COVID-19 pandemic, offering strategies for managing these disruptions. These influential works highlight significant advancements and shifts in the field, particularly the move towards real-time data collection, analysis, and automation in agricultural and food supply chains.

It is important to note that the slight decline observed in the graph from 2022 to 2024 should not be interpreted as a waning in interest. Our search was conducted on 25 June 2024, and not all research conducted during the examined period had been published in the Web of Science database at that time.

3.1.2. Most Globally and Locally Cited Documents

Figure 4 presents a chart displaying the papers with the highest numbers of global citations. Notably, Kusiak, 2018 [35] discusses the concept of the digitization of the supply chain as part of smart manufacturing through the application of IoT, cloud computing, service-oriented computing, artificial intelligence, and data science. Wang et al., 2018 [36] reviews the use of IoT, social networking, and crowdsourcing in supply chain operations and proposes machine learning models to process the large volumes of generated data. Liu et al., 2021 [30] examines how emerging technologies such as IoT, robotics, artificial intelligence, big data analytics, and blockchains improve the agri-food supply chain. Tang et al., 2019 [37] explores the strategic role of logistics and transportation services in creating economic, environmental, and social values through the adoption of new technologies. Saurabh et al., 2021 [32] studies the application of emerging information and communication technology in improving agri-food supply chain efficiency and quality management. Misra et al., 2022 [31] examines the role of IoT, big data, and artificial intelligence in shaping the future of agri-food systems and modernizing supply chains. These papers highlight the versatility of IoT and AI applications across supply chains in various industries. These technologies find utility in sectors ranging from agricultural supply chains to manufacturing supply chains and beyond.

In terms of local citations, the paper by Tsang et al., 2018 [38], holds the highest ranking with eight citations. This paper proposes an IoT- and AI-based risk monitoring system (IoTRMS) for controlling product quality and occupational safety risks in cold chains. Following this, the paper by Misra et al., 2022 [31], ranks second in the list of most locally cited references, as depicted in Figure 5. The papers by Tang et al., 2019 [37]; Saurabh et al., 2021 [32]; Hopkins et al., 2018 [39]; Hassija et al., 2021 [40]; Bhargava et al., 2022 [41]; and Alfian et al., 2020 [42] follow next on the list, each with five citations. Hopkins et al., 2018 [39] documents the role and impact of big data analytics and IoT in supporting a large logistics firm’s strategy to improve drivers’ safety, lower operating costs, and reduce the environmental impact of their vehicles. Hassija et al., 2021 [40] discusses the application of emerging technologies, such as blockchains, ML, and physically unclonable functions (PUFs), to address vulnerabilities in supply chain infrastructure. Bhargava et al., 2022 [41] presents an Industrial Internet of Things model integrated with intelligent logistics and transportation management to optimize logistics, improve customer experience, and minimize transportation costs. Alfian et al., 2020 [42] discusses a food traceability system that utilizes RFID and IoT sensors for identifying and tracking perishable items, such as meat and dairy products, to safeguard safety and quality.

The results of the local and global citation analysis reveal that issues such as smart manufacturing and mobile edge computing extend beyond the IoT-AI-based supply chain ecosystem, attracting attention from broader research areas. This highlights how advancements in these fields can create valuable spin-offs, potentially enhancing research across various domains. Conversely, within the IoT-AI supply chain ecosystem, applications focused on advancing agriculture and the food industry, as well as specialized risk monitoring systems for cold chains, address specific regional needs. These topics are highly relevant to local researchers and practitioners, underscoring their importance in improving agricultural practices and managing niche risks within particular regional contexts.

3.1.3. Most Relevant Keywords

Figure 6 displays a tree map illustrating the top 25 most frequently used authors’ keywords, generated through Biblioshiny. A tree map visually indicates the distribution and occurrence of keywords utilized by various authors in their publications. In this representation, the size of each segment, or “leaf”, in the tree map corresponds to the frequency of a specific keyword, as outlined by [43].

The terms “internet of things”, “supply chain”, “logistics”, “machine learning”, “artificial intelligence”, and “deep learning” were among the keywords in our main query, so they exhibit high frequency values in the tree map figure.

Notably, keywords such as “blockchain technology”, “industry 4.0”, “big data”, “security”, “edge computing”, “cloud computing”, “anomaly detection”, and “sustainability” emerge prominently. Blockchain technology is a shared, immutable ledger that facilitates the process of recording transactions and tracking assets in a business network [44]. Recently, blockchain technology has gained considerable attention from researchers and practitioners due to its unique features, including decentralization, security, reliability, and data integrity [45]. In this regard, [46] analyzes the inherent challenges and opportunities within IoT systems and the transformative capabilities of blockchains and proposes a conceptual architecture as a foundation for future research.

Security, privacy, and cybersecurity are among the main challenges of integrating IoT into supply chains. On the other hand, terms like data models, federated learning, and anomaly detection relate to AI methodologies. The simultaneous appearance of these two groups of keywords in the tree map suggests a growing interest in exploring machine learning-based solutions for advancing IoT in supply chains. Looking at the keywords from another perspective highlights terms such as COVID-19 and sustainability, indicating that the combination of IoT and AI not only enhances supply chain sustainability but also strengthens its resilience during disruptions like the COVID-19 pandemic.

3.1.4. Trend Topic Analysis

The trend topic analysis diagram in Figure 7, created using Biblioshiny, visually illustrates the progression and significance of research themes within a specific academic domain across time. The diagram utilizes a horizontal axis to denote the timeline, presenting years in chronological order. On the vertical axis, various subjects or terms are represented as bubbles or nodes [47], facilitating the observation of topic emergence and evolution over time. The size of each bubble indicates the relative importance or frequency of a given subject, with larger bubbles denoting more influential or extensively researched topics. While the tree map provides a broad overview of the most frequently occurring keywords during the selected time span, the trend topic analysis offers deeper insights into the dynamics of keyword co-occurrence over time.

Figure 7 represents an examination of the current trending subjects based on the authors’ keywords extracted from the dataset. During the analysis, the timespan was defined from 2011 to 2024, with a minimum word frequency set at five and an average of three words per year considered. The outcome of the trend topic analysis highlights “federated learning” and “digital twin” as the most recent and highly popular topics in the intersection of IoT and AI within the realm of supply chain management. This analysis, as depicted in Figure 7, emphasizes the growing interest in these areas, alongside enduring subjects like “smart manufacturing” and “cyber-physical systems”.

A substantial body of literature is dedicated to “federated learning”, which has emerged as a leading solution for training machine learning models within the context of distributed devices, using cutting-edge technology and expanding data repositories originating from IoT [48]. Federated learning is a secure machine learning technology proposed to protect data privacy and security in machine learning model training [49]. Federated learning has become increasingly attractive in the areas of wireless communications and machine learning due to its powerful learning ability and potential applications [50]. In recent years, federated learning has received widespread attention as a technology to solve the problem of data islands, and it has begun to be applied in fields such as finance, healthcare, and smart cities [51]. Prime decentralized technologies such as blockchains and federated learning are instrumental in tackling the challenges faced by IoT in terms of centralization, privacy protection, latency, and security [52]. Meanwhile, a “digital twin” represents a virtual replica of a tangible product or process in the physical world, constructed using data and real-time information gathered through sensors, effectively simulating the behavior and performance of a real-world manufacturing environment [53].

Industry 4.0 technologies, such as “cyber-physical systems”, “big data”, “internet of things”, “digital twins”, and “artificial intelligence”, play a crucial role in managing the complexities of producing highly personalized products at scale. They necessitate the digitization and fusion of information and operational technologies for agile and automated manufacturing processes [54]. Cyber–physical systems, also known as smart systems, closely integrate software and physical components, offering diverse applications across various industries [55].

The term “smart manufacturing” surfaced in 2018 and remained prevalent until 2023. Smart manufacturing combines current and future manufacturing assets with sensors, computing platforms, communication technologies, data-driven modeling, control mechanisms, simulation, and predictive engineering [35].

The results of the trend topic analysis not only confirm the findings from the tree map analysis but also provide new perspectives for future research. The recent rise of federated learning in the literature highlights the growing interest in leveraging machine learning techniques to address key challenges in IoT integration within supply chains, particularly around privacy, security, and data isolation. This underscores the significance of privacy and security as major concerns when applying IoT to supply chains. Additionally, areas such as smart cities and smart manufacturing are likely to face similar challenges. The analysis suggests that while smart manufacturing and cyber–physical systems remain central themes, newer concepts like digital twins and federated learning are gaining significant traction, indicating a shift toward more secure and decentralized technologies.

3.1.5. Co-Occurrence Network of Keywords

This section helps answer research question RQ 2, i.e., “What IoT strengths and weaknesses affect AI/ML positively or negatively in the studies?” Network visualization and co-occurrence network analysis have emerged as crucial subjects in the current academic landscape. Through these methods, researchers can evaluate the relationships among various entities, facilitating a comprehensive analysis of the connections between them [56]. The co-occurrence network of the authors’ keywords serves as a visual tool that interprets the interrelations among keywords utilized by authors to delineate their research findings. Keywords are deemed to co-occur if they are present in the same title, abstract, or citation context [57]. This network was constructed using VOSviewer, which assigns nodes to represent keywords and establishes connections between them based on their co-occurrence in publications. To ensure the network’s relevance, a minimum occurrence threshold of five was set, meaning that only keywords appearing in at least five distinct publications were included in the analysis. This criterion filtered out less common terms, emphasizing frequently used keywords and enhancing the network’s clarity and significance, resulting in a final count of 77 keywords, which was deemed optimal for visualizing well-defined clusters and identifying key trends. When the threshold was increased to 10 or more, the network diagram became significantly smaller, with smaller cluster sizes. Thus, the threshold of five was chosen to generate a network diagram with well-defined clusters and an ideal number of keywords.

In a network, the size of each node corresponds to the frequency of the keyword’s occurrence. Larger nodes indicate keywords that appear more frequently within the dataset. VOSviewer employs colors to represent different clusters or themes within a network, grouping closely related or frequently co-occurring keywords together.

By analyzing the dataset of 810 documents in VOSviewer, we identified seven clusters in the co-occurrence network diagram of the authors’ keywords, as depicted in Figure 8.

The red cluster is primarily centered around keywords such as “supply chain”, “industry 4.0”, and “digital technologies”, indicating a focus on the digitalization of the supply chain through the application of Industry 4.0 technologies. The evolution of industry and technology has progressed rapidly through several revolutionary phases, each bringing significant advancements in SCM processes. Ref. [58] outlines the progression from the First Industrial Revolution, marked by the steam engine to Industry 4.0, characterized by the convergence of physical, digital, and biological spheres. Beyond Industry 4.0, the emergence of Industry 5.0 is notable, emphasizing personalized manufacturing and human–machine collaboration. Ref. [59] identifies crucial enablers of Industry 5.0, including AI, big data, digital transformation, ML, and IoT, reshaping sectors such as manufacturing and supply chains. Ref. [60] highlights how IoT, AI, robotics, and advanced computing challenge traditional approaches in pharmaceutical production. In the post-COVID-19 era, there is a growing emphasis on leveraging these technologies for socially responsible operations. Ref. [61] proposes a framework that integrates Industry 4.0 technologies with socially responsible operational performance, identifying seven transformational technologies to enhance organizational and community social performance. Additionally, the circular economy (CE) gains importance in post-pandemic recovery. Ref. [62] suggests evaluating CE implementation through “non-market-based environmental goods valuation methods” and improving performance measurement using a digital technology ecosystem involving multiple stakeholders. This integration of advanced technologies with sustainability and social responsibility signifies the next phase in industrial evolution, promising more efficient, flexible, and responsible supply chain operations.

The frequent occurrence of keywords in the green cluster, such as “healthcare”, “optimization”, and “monitoring”, suggests a comprehensive approach to improving the effectiveness of healthcare supply chains. Challenges related to the perishability and strict time constraints of healthcare products, such as blood products, are among the key issues that recent research is actively working to address [63]. The fusion of IoT and healthcare technologies is reshaping healthcare systems. Ref. [64] emphasizes the crucial role of IoT in combating pandemics like COVID-19 through intelligent monitoring and remote patient management. The authors’ proposed cloud-enabled IoT framework not only aids in pandemic control but also establishes a basis for resilient smart communities. Ref. [65] tackles data privacy issues in healthcare by introducing federated learning (FL) as a solution. The authors’ deep FL framework ensures data privacy, boosts model accuracy, and reduces operational costs, showing promise for healthcare data monitoring. Ref. [66] presents a privacy-preserved smartphone-based personal assistant (PP-SPA), a privacy-preserving framework for human activity recognition tailored for cognitively impaired individuals. By using smartphone sensors and ML models, PP-SPA provides personalized health assistance with an impressive accuracy rate of 90%. Ref. [67] explores advancements in personalized healthcare using wearable devices and real-time machine learning on edge computing technologies. The authors’ proposed architecture improves efficiency, privacy, and timely health event predictions, laying the groundwork for scalable healthcare systems. Ref. [68] investigates the fusion of IoT and omnichannel services in healthcare, focusing on the importance of wearables in connecting providers and patients. The authors’ data-driven analytical model aims to streamline healthcare supply chain management, addressing ethical concerns and enhancing customer engagement. This cluster highlights the transformative potential of IoT and data-driven applications in reshaping the healthcare ecosystem, with the healthcare supply chain serving as a critical element. Emphasizing privacy, efficiency, and personalized care, these innovations promise to enhance the resilience and effectiveness of healthcare systems, including supply chain operations.

Turning to agriculture, the blue cluster emphasizes smart agriculture with keywords such as “internet of things”, “deep learning”, “sensors”, and “smart agriculture”. The convergence of agriculture and technology is undergoing a significant transformation. Ref. [69] highlights the essential role of IoT and blockchains in strengthening the food supply chain, emphasizing the incorporation of advanced deep learning techniques for improved efficiency. The authors’ hybrid model, which combines long short-term memory (LSTM) and gated recurrent units (GRUs) with genetic algorithm (GA) optimization, demonstrates the potential to revolutionize supply chain management through predictive analytics. Ref. [70] highlights precision agriculture, introducing a deep learning neural network-based IoT-enabled intelligent irrigation system for precision agriculture (DLiSA), an IoT-enabled intelligent irrigation system empowered by deep learning neural networks. By forecasting soil moisture content and optimizing water distribution, DLiSA emerges as a sustainable solution for maximizing agricultural yields within resource constraints. Ref. [71] envisions a future where smart agriculture is crucial for meeting rising food demands. The authors’ exploration of smart technology integration, encompassing IoT, AI, and blockchains, underscores the transformative capacity of precision agriculture in guaranteeing food security and sustainability. Ref. [72] presents an innovative anomaly detection model based on generative adversarial networks (GANs) for smart agricultural IoT data, enhancing data analysis and decision-making in agricultural management. Finally, Ref. [73] suggests a comprehensive system that merges deep learning, blockchains, and IoT for greenhouse operations, demonstrating how technology can optimize crop production, ensure traceability in supply chain, and promote sustainable agricultural practices. These research initiatives collectively illuminate a path towards smarter, more efficient, and resilient agricultural systems.

The yellow cluster, marked by frequent keywords such as “industrial internet of things”, “digital twins”, “cloud computing”, “manufacturing”, and “big data analytics”, explores the domain of smart manufacturing. Industry 4.0, driven by the IoT and cyber–physical systems, brought in a new era of intelligent manufacturing environments, where customized smart products evolve from sophisticated processes. Ref. [74] highlights the transformative potential of cyber–physical systems in revolutionizing production and logistics, emphasizing the integration of big data capabilities and predictive decision support tools to enhance productivity and efficiency. Ref. [75] explores the significance of the digital twin as a cornerstone of Industry 4.0, facilitating simulation and optimization within a complex technological ecosystem while emphasizing the need to address security threats. Ref. [76] focuses on asset management in Industry 4.0, promoting predictive maintenance through genetic algorithm-based resource management integrated with machine learning, demonstrating superior performance in terms of execution time, cost, and energy consumption. Ref. [77] addresses data security challenges in industrial federated learning, introducing a federated group synchronization framework (FEDGS), a hierarchical cloud–edge–end framework tailored for 5G industries to enhance federated learning performance amid non-independent and identically distributed data, exhibiting improved accuracy and efficiency. Lastly, ref. [78] presents a smart system that leverages IoT and digital twin technologies for real-time occupational safety monitoring in warehouses, employing deep neural networks for anomaly detection and bluetooth low energy for indoor localization, providing a comprehensive solution for ensuring workplace safety and efficiency. These studies collectively shed light on the diverse landscape of Industry 4.0, emphasizing the pivotal role of advanced technologies in reshaping industrial paradigms towards intelligent, secure, and optimized operations.

The violet cluster, characterized by keywords such as “big data”, “edge computing”, “cybersecurity”, “smart city”, and “cyber-physical systems”, shifts attention to cybersecurity within smart cities. The advancement of computing paradigms, propelled by the IoT and cloud services, has spurred the rise of edge computing. This facilitates data processing at the network’s edge to meet response time, energy efficiency, and privacy requirements [79]. In the field of smart cities, ref. [80] stresses the importance of integrating big data analysis and deep learning algorithms to improve governance and data processing efficiency. The authors’ study demonstrates the effectiveness of a distributed parallelism strategy and digital twin technology, showcasing notable enhancements in energy efficiency and data transmission performance. Ref. [81] emphasizes the crucial role of IoT in shaping smart cities, particularly in managing vital resources like water through machine learning techniques for forecasting and anomaly detection. Ref. [82] explores the challenges associated with centralized cloud-based data analytics and proposes a distributed learning framework for edge computing, highlighting the advantages of conducting analytics closer to data sources. Ref. [83] expands the discourse to encompass the comprehensive integration of digital technologies in smart cities, emphasizing the transformative effects across various sectors while recognizing persistent challenges that hinder progress. The author offers insightful counter-measures and recommendations for overcoming these obstacles. Collectively, these studies underline the transformative potential of advanced technologies in reshaping urban landscapes towards intelligent, efficient, and sustainable futures.

In the cyan cluster, prominent keywords such as “machine learning”, “smart grids”, “smart meters”, and “RFID” signal a focus on applying machine learning within smart grids. The merging of state-of-the-art technologies in energy, logistics, power grids, and smart grids is reshaping urban environments and infrastructure. Ref. [84] presents an innovative federated learning-based AI of things scheme for secure energy data sharing in smart grids, emphasizing communication efficiency and privacy preservation through edge–cloud collaboration. Ref. [85] illuminates the digital transformation of port logistics, stressing the necessity of data-driven approaches for sustainable development. The authors utilize various advanced technologies to optimize flow management and stakeholder engagement. Ref. [86] explores power grid control strategies, concentrating on distributed energy storage, electric vehicle charging technologies, and cyber–physical security measures to ensure stability and resilience in smart grids, thereby paving the way for a robust smart distribution grid. Similarly, ref. [87] addresses the growing threat of electricity theft in modernized power systems, advocating for comprehensive detection methods employing ML and measurement mismatch techniques to safeguard against adversarial attacks. Lastly, ref. [88] explains the crucial role of fog computing in enhancing the efficiency of smart electrical grids by introducing a novel three-tier architecture and a sophisticated electrical load forecasting strategy. They underscore the significance of feature selection methodologies, such as fuzzy-based feature selection, for precise and prompt decision-making processes. Collectively, these research efforts underscore the transformative potential of advanced technologies in reshaping urban environments towards intelligent, efficient, and sustainable futures.

The orange cluster, characterized by keywords like “security”, “intrusion detection systems”, “neural network”, “authentication”, and “energy management”, focuses on security within energy management systems. The incorporation of state-of-the-art technologies, such as the IoT, AI, and ML, is propelling significant progress in smart energy systems and grid management.

Altogether, the seven clusters focus on the interdependence of IoT and AI. These clusters address research question RQ 2 by highlighting the strengths and weaknesses of IoT that impact AI systems. IoT technologies enable the collection of extensive data from diverse sources, such as sensors, devices, and systems, providing rich datasets for AI and ML algorithms. Real-time data collection facilitated by IoT allows AI models to make prompt decisions and predictions. Automation through IoT devices streamlines data collection and processing, enhancing the efficiency of AI systems. The seamless connectivity promoted by IoT enables AI models to access and analyze data from multiple sources, supporting predictive analytics and informed decision-making. Additionally, IoT optimization of operations and processes enhances the performance of AI algorithms across supply chains in healthcare, agriculture, manufacturing, and energy sectors. However, the substantial data volume generated by IoT devices raises concerns regarding data security and privacy, potentially exposing sensitive information to cyber threats. Variances in data quality and consistency within IoT data pose challenges for data preprocessing and model training in AI applications. Compatibility issues among IoT devices and platforms can impede data integration and the smooth operation of AI systems. Scaling IoT infrastructure to manage large data volumes and devices may present hurdles in efficiently processing data for AI models. The complexity of IoT ecosystems, especially in smart city environments, can complicate ensuring the reliability and robustness of AI solutions. Moreover, energy consumption by IoT devices necessitates efficient power resource management for sustainable AI applications within IoT environments.

These strengths and weaknesses underscore the intricate relationship between IoT technologies and AI applications, emphasizing the importance of developing robust strategies to leverage IoT strengths while addressing its weaknesses for effective AI integration.

3.1.6. Co-Citation Network of Cited References

This section provides the answer to research question RQ 3, i.e., “What AI/ML strengths and weaknesses affect IoT positively or negatively in the studies?” The term “co-citation” refers to two articles that are both referenced in another article [89]. The larger the number of publications by which two publications are co-cited, the stronger the co-citation relation between the two publications [90]. The co-citation network diagram in Figure 9 illustrates seven clusters at the intersection of IoT, supply chains, and AI, organized according to a co-citation analysis. Based on our analysis, the results reveal an overlap between five clusters in keyword co-occurrence and the co-citation network. To maintain focus, we expanded on two newly identified clusters in the co-citation analysis: machine learning and optimization, and IoT-related clusters. We provided only a general overview of the previously discussed clusters to keep the study concise.

The red cluster highlights various advancements in machine learning and optimization algorithms. Ref. [91] introduces federated learning, a decentralized approach that collects locally computed updates to train models without centralizing sensitive data. This method overcomes challenges related to data privacy and quantity. Ref. [92] presents the Adam algorithm, an efficient optimization method for stochastic objective functions that adapts to varying gradients, demonstrating practical effectiveness and convergence properties. Refs. [93,94] highlight the efficacy of deep learning and neural networks in various applications like character recognition and data representation learning, emphasizing the success of convolutional neural networks in image processing and the ability of deep learning models to understand complex structures in large datasets, improving tasks such as speech recognition and object detection. Ref. [95] addresses security challenges in IoT systems, suggesting the bot–IoT dataset for training intrusion detection systems and stressing the importance of realistic datasets for effective system development and evaluation in the face of evolving cyber threats. Ref. [96] discusses the significance of big data and digital twin technologies in advancing smart manufacturing, underlining the role of data intelligence and cyber–physical integration in improving various manufacturing processes, from product design to predictive maintenance. The authors highlight the integration of big data analytics and digital twins as a key strategy for promoting smart manufacturing. The papers in this cluster collectively showcase ongoing research and development in machine learning, optimization algorithms, deep learning, IoT security, and smart manufacturing, emphasizing the importance of innovative approaches in tackling contemporary challenges and propelling technological advancements in diverse domains.

The research in the green cluster explores the field of IoT, showcasing its impact and potential across various domains. An essential aspect discussed is the integration of technologies and communication solutions to advance IoT applications. Ref. [97] stresses the necessity of synergy across telecommunications, informatics, electronics, and social science for IoT development. Ref. [98] elaborates on how wireless sensor network technologies underpin the IoT, enabling ubiquitous sensing and establishing a network where sensors and actuators seamlessly engage with the environment. The authors consider IoT a ground-breaking technology that will revolutionize the internet into a fully integrated platform. Ref. [99] underlines the significance of smart sensors, communication technologies, and internet protocols in advancing IoT, with the goal of connecting physical objects for intelligent decision-making. Ref. [100] explores industrial IoT applications, focusing on the role of RFID, wireless, mobile, and sensor devices in constructing robust systems. The authors methodically examine key enabling technologies and current applications, outlining research trends and challenges. Ref. [101] provides a comprehensive overview of IoT’s influence on SCM, categorizing the literature by methodology, industry sector, and supply chain processes. The authors stress the necessity for more analytical models and empirical studies to grasp IoT’s implications for SCM, especially in delivery supply chains and food and manufacturing sectors. Collectively, these papers highlight the evolving landscape of IoT technologies and applications, stressing the significance of interdisciplinary collaboration and innovative approaches in shaping the future of IoT implementation and research.

Moving from IoT to the agricultural domain, the blue cluster provides a comprehensive insight into the evolving landscape of technology within agriculture and food supply chains. The central theme highlights the utilization of advanced technologies, such as big data, blockchains, and IoT, to improve traceability, safety, and efficiency in the agri-food sector.

The yellow cluster explores technology integration in supply chain management. The overarching themes in this cluster center on harnessing advanced technologies to refine supply chain management, amplify visibility, and mitigate disruption risks in the dynamic realm of Industry 4.0.

The violet cluster emphasizes the transformative potential of Industry 4.0 and its implications for logistics, supply chain management, and sustainability. Industry 4.0, also known as the Fourth Industrial Revolution, is reshaping industries through cutting-edge technologies such as big data analytics, IoT, and interconnected systems. The papers in this cluster indicate a shift towards more connected, data-driven, and sustainable supply chains, encouraging organizations to adopt technological advancements to maintain competitiveness and resilience in the evolving business landscape.

The cyan cluster explores the potential of blockchain technology in revolutionizing supply chain management by enhancing transparency, traceability, and efficiency. The studies underscore blockchains’ transformative potential in supply chain management, promoting increased transparency, efficiency, and trustworthiness.

Finally, the orange cluster explains the fusion of digital manufacturing and Industry 4.0 principles within the domain of supply chain management. It provides valuable insights into fortifying resilience, refining decision-making processes, and optimizing performance within intricate supply chain networks.

Collectively, all seven clusters respond to research question RQ 3 by delineating the strengths and weaknesses of AI impacting IoT. The red cluster discusses federated learning, which stands out as a decentralized training approach that sidesteps centralizing sensitive data, thereby tackling data privacy and quantity hurdles in IoT applications. The Adam algorithm emerges as an efficient optimization technique catering to stochastic objective functions, adjusting to varying gradients, showcasing practical efficacy, and convergence properties in IoT optimization. Furthermore, the efficacy of AI in various applications such as image processing, character recognition, speech recognition, and object detection enhances IoT tasks that necessitate an understanding of intricate structures within vast datasets. The study also addresses security challenges within IoT systems by utilizing datasets like Bot-IoT and UNSW-NB15, underlining the significance of authentic datasets for efficient system development and evaluation. Moreover, the importance of big data and digital twin technologies is underscored in propelling smart manufacturing, amalgamating data intelligence with cyber–physical systems, and fostering IoT applications across manufacturing processes from design to predictive maintenance. Despite federated learning’s strides in mitigating data privacy challenges, ensuring the absolute privacy and security of shared data in IoT systems remains a pressing concern. Additionally, the complexity and interpretability limitations of deep learning models pose challenges in comprehending their decision-making processes within IoT applications. The implementation of AI/ML algorithms, particularly deep learning, in IoT devices may necessitate substantial computational resources and energy, impacting the scalability and efficiency of IoT systems. While endeavors to bolster security in IoT systems persist, they are still vulnerable to cyber threats, mandating continual updates and robust security measures to counter risks tied to AI/ML integration. Ensuring the quality and unbiased nature of data utilized in AI/ML algorithms for IoT applications is imperative to avert biased decision-making and inaccurate outcomes.

In summary, these strengths and weaknesses bring into the limelight the complex interplay between AI/ML advancements and IoT applications, emphasizing the necessity for sustained innovation and diligence in surmounting challenges to harness the full potential of these technologies across diverse domains.

3.2. Topic Modeling and Cluster Identification

This section provides the answer to our second research question, RQ 4, regarding how AI/ML and IoT can address the challenges of digitized supply chains. It achieves this by discussing the results obtained after performing topic modeling on the dataset of 810 documents using the NMF algorithm. The topic modeling results also reveal overlaps in five topics with the clusters from the keyword co-occurrence and co-citation analysis. Consistent with our previous approach, we provide a detailed exploration of the novel topics, including cold-chain logistics and water distribution systems, and limit the discussion of repetitive clusters to a brief overview.

Figure 10 displays our topic modeling outputs for the NMF technique visualized using t-SNE. Each data point in the figure represents a scientific publication from the dataset [102]. T-SNE is a visualization technique for high-dimensional data, assigning each data point a position on a two or three-dimensional map [103]. It is an improved version of Stochastic Neighbor Embedding [104], offering enhanced visualizations by preventing points from clustering in the map’s center. A clear comparison between the t-SNE graphs of NMF and LDA reveals the superior performance of NMF in topic modeling over LDA. Particularly noteworthy was the reduced overlap and increased prominence of clusters in the NMF analysis. Consequently, we opted for the topic modeling results derived from the NMF algorithm. When applied to NMF topics specifically, t-SNE clustering facilitates the exploration of relationships and similarities among the topics. By reducing the dimensionality of the topic representations derived from NMF, t-SNE creates a visual representation that groups together topics with similar characteristics in close proximity.

To determine the optimal number of topics, we employed a data-driven approach by evaluating model perplexity and topic coherence metrics. Model perplexity assesses how well a model generalizes to unseen data, while topic coherence measures the semantic consistency of high-ranking words within each topic. We also used an inter-topic distance map to visualize topic separation and overlap, ensuring that the selected topics were well-defined and meaningful.

Our iterative tuning process showed improved topic coherence and reduced model complexity as the number of topics increased. However, the optimal balance was reached at seven topics, where the inter-topic distance map displayed large, well-separated clusters with minimal overlap. The selected structure ensures that each topic represents a distinct research theme in AI-driven CRM, covering core discussions in customer analytics, predictive modeling, automation, and AI ethics. Figure 10 illustrates the topic distribution, where the circle sizes reflect topic prevalence, and their spatial positioning is determined by multidimensional scaling.

Figure 11 presents an inter-topic distance map that visualizes topics on the left and the top 30 most frequent terms on the right. This visual chart provides valuable insights into the topics. Each of the seven topics is represented by a bubble, with the size indicating its relevance in the corpus. Larger bubbles indicate more relevant topics. Furthermore, topics that are closer on the plot are more similar to each other compared to those that are farther apart. Our topic model has fairly big, non-overlapping bubbles scattered throughout the chart instead of being clustered in one quadrant, which signifies that our topic model is of high quality [105].

In our analysis, Topic 1, labeled as “AI and Machine Learning”, stands out with the largest bubble size, signifying its prominence and significant composition of the top terms, accounting for 28.5% of the tokens. None of the topics show overlap, indicating distinctiveness between them. While terms like “iot”, “machine_learning”, and “deep_learning” are frequently present as expected, we can also observe the occurrence of terms such as “attack”, “security”, and “performance” among the most common ones. Thus, the papers on this topic deal with the challenge of data security and cyber-attacks in IoT-enabled supply chains.

Figure 12 offers insights into the seven topics derived through the NMF algorithm. It showcases word clouds that visually represent each topic, highlighting the top 25 frequently used words.

Topic 1 focuses on utilizing machine learning to address data security challenges, as indicated by the prevalence of terms such as “data”, “model”, “algorithm”, “machine learning”, “attack”, and “security”. Given the vulnerability of IoT devices to cyber-attacks, safeguarding sensitive data is crucial. This topic discusses innovative approaches to tackle critical challenges in IoT data management, federated learning, intrusion detection, and cybersecurity within evolving networks like SC 4.0.

Topic 2 explores the theme of digitalization of agri-food supply chains. It examines the transformative potential of IoT, big data, and AI in agri-food systems, emphasizing the importance of sensor data, social media insights, and advanced technologies in modernizing agriculture and food processes.

Topic 3, known as cold-chain logistics, explores the evolution of logistics in the digital age, creating transformative changes across various sectors. Ref. [106] advocates for enhancing cold-chain logistics by integrating big data, artificial intelligence, and the Energy Internet of Things. Their proposed “four-in-one” cold-chain system enables real-time monitoring and predictive analysis and fosters data sharing to optimize resource allocation and drive cold-chain logistics towards low-carbon green development. Ref. [107] highlights the booming e-commerce landscape’s impact on logistics demands, emphasizing the need for optimized node layouts to enhance efficiency and reduce costs. By leveraging IoT, big data, and deep learning algorithms, the author’s study introduces an improved logistics node layout scheme, addressing transportation costs, distribution efficiency, and information accuracy to meet evolving consumer needs effectively. Ref. [108] stresses the importance of modernizing the logistics industry to curb costs and elevate management standards. The authors propose a modern logistics system characterized by informatization, intelligence, and traceability, interconnecting various systems and networks to streamline operations. By integrating technologies like big data, IoT, cloud computing, and AI, their framework lays the groundwork for a more efficient and interconnected logistics ecosystem. Ref. [109] explores marine logistics, advocating for a shared information platform powered by IoT and cloud computing. Their research highlights real-time monitoring and cargo positioning, significantly enhancing the efficiency of marine logistics management. Ref. [110] introduces Logistics Industry 4.0, showcasing how emerging technologies like IoT, big data, AI, and blockchains can revolutionize supply chain processes. The authors’ exploration of how these technologies transform transportation, warehousing, production, and supply chain operations underscores the vast potential and challenges in leveraging technology to reshape the logistics landscape. Ref. [111] explores how ML techniques, including neural networks and random forests, improve sales forecasting in the food industry and lead to more accurate predictions.

Topic 4, focusing on smart grids and energy management systems, underscores the significance of smart grids in meeting rising energy demands. The recent studies on technology integration in energy systems emphasize the crucial role of advanced solutions like smart energy management systems, IoT applications, and modern logistics technologies in transforming industries towards improved efficiency, security, and sustainability.

Topic 5 is labeled as “supply chain performance”. This topic underscores the pivotal role of IoT, cloud computing, and advanced technologies in revolutionizing supply chain management and risk assessment, thereby enhancing efficiency, security, and competitiveness in these critical domains.

Topic 6, titled “water distribution systems”, includes research on smart water meters and distribution systems with the aim of minimizing losses and enhancing efficiency, as indicated by keywords such as “water”, “water distribution”, “irrigation”, “monitoring”, “sensor”, “consumption”, “leakage”, and “urban”. Efficient water management in urban areas, crucial due to factors like population growth and climate change, requires innovative solutions. Ref. [112] introduces the Water Wise System (W2S), utilizing digital water solutions and IoT to revolutionize water distribution network management. By incorporating ML and deep learning, this system aims to predict and analyze water challenges, integrating technologies like SCADA (Supervisory Control and Data Acquisition) and GISs (Geographic Information Systems) for a smarter approach. Ref. [113] proposes a smart water management system utilizing the ZR16S08 microcontroller and Raspberry Pi for real-time water pipe monitoring, emphasizing the importance of quality water supply and loss prevention. Ref. [114] highlights the importance of implementing smart water management systems in parks to conserve water resources efficiently. The authors’ multidisciplinary approach combines IoT, sensor technologies, and machine learning to optimize irrigation, reducing water wastage and nutrient seepage. Ref. [115] presents an IoT-based framework for real-time water quality management, focusing on proactive monitoring and machine learning analysis for efficient water quality control. Ref. [116] stresses the urgency of real-time water quality monitoring systems to safeguard public health, utilizing IoT for faster processing and active contamination reduction, thereby emphasizing the critical role of technology in effectively addressing water quality challenges. Collectively, these studies advocate for the integration of IoT, ML, and smart technologies in water management systems to enhance efficiency, optimize resource usage, and ensure environmental sustainability.

Topic 7 is labeled as “smart manufacturing”. The manufacturing industry is undergoing a significant transformation towards smart manufacturing, integrating advanced technologies like IoT, AI, and data science. This topic illuminates the trajectory towards intelligent manufacturing environments, emphasizing the integration of cutting-edge technologies and strategies to meet evolving industrial demands and challenges.

Collectively, all the seven topics help to answer research question RQ 4 by addressing key challenges within supply chains, with a particular emphasis on data security, digitalization, and logistics optimization. A significant hurdle involves ensuring data security in IoT-driven supply chains, where machine learning plays a crucial role in managing privacy and security risks. Given the susceptibility of IoT devices to cyber-attacks, innovative strategies are essential to handle IoT data securely, improve intrusion detection, and safeguard cybersecurity within evolving networks like SC 4.0. Solutions range from developing distributed indexing mechanisms to crafting personalized federated learning models and employing ensemble-based intrusion detection systems to counter intricate cyber threats. Digitalizing agri-food supply chains presents another complex challenge, with Industry 4.0 technologies—such as IoT, big data, and AI—reshaping supply chain dynamics by bolstering resilience and visibility. These technologies play a pivotal role in modernizing agricultural and food processes, enhancing food production, quality assessment, and safety standards. Additionally, the studies explore the integration of Industry 4.0 technologies in sustainable supply chain management, advocating for sustainability through a blend of IoT, cloud computing, big data, AI, blockchains, and digital twins. Cold-chain logistics emerges as a critical focal point, with researchers advocating for the fusion of big data, AI, and the Energy Internet of Things to bolster real-time monitoring, predictive analysis, and resource optimization. This strategic approach aims to steer cold-chain logistics along a low-carbon, environmentally sustainable trajectory, addressing challenges stemming from the expanding e-commerce landscape and the necessity for efficient logistics node layouts. Collectively, these studies underscore the transformative potential of cutting-edge technologies in tackling crucial supply chain challenges across diverse sectors.

Table 1. Summarized results of bibliometric analysis and topic modeling.

Theme	Co-Occurrence Network Analysis of Authors’ Keywords			Co-Citation Network Analysis of Cited References				Topic Modeling
	Keyword	O	TLS	Document	Citations	Source	TLS	Keywords
AI and Machine Learning				(McMahan et al., 2017)	15	Proceedings of Machine Learning Research	29	data
				(Kingma & Ba, 2014)	11	Computer Science & Engineering	24	machine learning
				(Lecun et al., 1998)	11	Proceedings of the IEEE	19	attack
				(LeCun et al., 2015)	11	Intelligent Control & Automation	23	security
				(Qi & Tao, 2018)	10	IEEE Access	49	network
Agri-Food Supply Chain	internet of things	317	666	(Wolfert et al., 2017)	11	Agricultural Systems	65	artificial intelligence
	deep learning	66	144	(Aung & Chang, 2014)	10	Food Control	53	food
	anomaly detection	20	50	(Lezoche et al., 2020)	5	Computers in Industry	50	technology
	sensors	14	51	(Kamath, 2018)	8	The Journal of the British Blockchain Association	65	big data
	smart agriculture	12	40	(Kamble et al., 2020)	8	International Journal of Information Management	59	blockchain
Smart Grid and Energy Management Systems	security	33	100					smart grid
	intrusion detection system	16	39					electricity
	neural network	13	19					distribution
	authentication	8	25					energy
	energy management	6	7					control
Smart Manufacturing	industrial internet of things	38	87	(Cavalcante et al., 2019)	6	International Conference on Service Systems and Service Management (ICSSSM)	33	manufacturing
	digital twins	23	49	(Oesterreich & Teuteberg, 2016)	5	Computers in Industry	39	smart
	cloud computing	22	86					production
	manufacturing	13	46					iiot
	big data analytics	11	29					industry
Digitalization of Supply Chain	artificial intelligence	118	307	(Saberi et al., 2019)	20	International Journal of Production Research	187	supply chain
	blockchain technology	87	252	(Ivanov et al., 2018)	18	International Journal of Production Research	177	management
	supply chain	86	200	(Hofmann & Rüsch, 2017)	16	Computers in Industry	125	risk
	industry 4.0	64	146	(Ivanov & Dolgui, 2021)	13	Production Plan Control	118	information
	logistics	28	68	(G. Wang et al., 2016)	13	International Journal of Production Economics	127	technology

(O: Occurrences, TLS: Total Link Strength).

presents the key themes arising from three distinct analyses of the dataset: AI and machine learning, the agri-food supply chain, smart grid and energy management systems, smart manufacturing, and digitalization of the supply chain. These thematic categories emerged from the co-occurrence network analysis of authors’ keywords, the co-citation network analysis of cited references, and topic modeling. Notably, the themes listed in the table recurred in at least two of the three analyses conducted, with the blank region indicating the non-occurrence of a theme in a particular analysis.

Conversely, specific topics like cold-chain logistics, smart water distribution systems, the optimization of healthcare, smart cities, and the Internet of Things made singular appearances in our analyses and, therefore, are absent from the tabulated results. The table further delineates the highly cited publications associated with each theme, alongside the prevalent keywords for each category.

3.3. Application of AI Algorithms in IoT-Based Supply Chain Management

This section addresses the research question RQ 5 by engaging in a detailed discussion of the various AI/ML algorithms used for addressing supply chain challenges. AI algorithms are crucial for enhancing supply chain management by improving operational efficiency, reducing costs, and enabling more informed decision-making. Each AI and ML technique offers unique capabilities, including forecasting, optimization, classification, clustering, and natural language processing. These capabilities are instrumental in addressing a wide range of challenges across supply chain operations.

Figure 13, generated from insights gathered in our analyses, illustrates the specific challenges in each module of the supply chain and highlights the AI-driven solutions identified across various research areas to address these challenges. The figure categorizes methodologies into four main types: supervised, unsupervised, reinforcement, and hybrid ML techniques. It is evident that supervised learning techniques are predominantly used to tackle many supply chain challenges. However, unsupervised, reinforcement, and hybrid methods are also effectively applied to address other key issues in the supply chain, offering diverse strategies for optimizing different processes.

Supervised learning algorithms, including decision trees, naïve Bayes, k-nearest neighbors (KNN), support vector machines (SVMs), and random forest, have become fundamental in tackling supply chain challenges, particularly those involving classification and regression tasks [117,118]. For instance, in agri-food supply chains, these algorithms are used to ensure product quality and detect pests. Random forest, in particular, has proven highly effective in intrusion detection systems within Industrial Internet of Things environments, outperforming other models like decision trees and SVMs [12]. This capability aligns well with the need for enhanced cybersecurity in modern supply chains, where data security is significant. In demand forecasting and dynamic pricing, artificial neural networks (ANNs) and gradient boosting are frequently used. ANNs, known for their ability to model complex relationships, assist in predicting demand trends, while gradient boosting improves the precision of demand forecasting [119]. These models allow organizations to make informed decisions, adjust pricing dynamically, and maintain optimal inventory levels [119]. Other supervised machine learning techniques, including deep learning models like convolutional neural networks (CNNs) and reinforcement learning (RL), are increasingly being applied in SCM to tackle more complex and dynamic challenges. CNNs, known for their layered architectures, have been successfully applied in supply chain risk prediction, especially during unforeseen disruptions like natural disasters. These models process vast amounts of data from IoT sensors, enabling real-time decision-making and proactive risk mitigation [120].

Unsupervised learning methods, such as k-means clustering and principal component analysis (PCA), are applied in scenarios where the goal is to group data or reduce dimensionality without labeled data. These techniques are particularly useful in customer segmentation, where understanding diverse customer preferences is crucial for targeted marketing strategies. Additionally, k-means clustering is utilized in supplier selection processes to categorize suppliers based on performance and risk factors, further refining decision-making capabilities [117].

Reinforcement learning is used in more dynamic applications, such as order replenishment management and emergency supply chain responses. It helps optimize collaboration between different nodes of the supply chain, ensuring timely replenishment and efficient management of resources during crises [119]. Reinforcement learning models dynamically learn from their environment, making them well-suited for managing supply chains in volatile conditions.

The integration of ML with other optimization methods, such as GAs and fuzzy methods, creates hybrid models that provide enhanced decision-making tools. For example, GAs are utilized for predictive maintenance, enabling organizations to make timely decisions regarding component repair or replacement. This prevents costly disruptions in production and maintains supply chain continuity [76]. GAs, which simulate natural evolutionary processes, are also used in demand forecasting by combining regression models to improve accuracy [121]. Fuzzy methods are another critical tool in handling uncertainties within supply chains. They are particularly useful in supplier selection, where multiple factors such as quality, cost, and risk must be weighed. By combining fuzzy logic with ML techniques, organizations can navigate dynamic supply chain configurations and make more flexible and accurate decisions [122].

The adoption of AI algorithms in supply chain management significantly advances operational excellence. From demand forecasting and dynamic pricing to cyber-attack prevention and supplier selection, AI-driven approaches streamline tasks, improve decision-making, and boost overall supply chain performance. The utilization of supervised learning for classification, unsupervised learning for clustering, deep learning for handling complex datasets, and hybrid methods for optimization demonstrates the versatility of AI in addressing diverse challenges.

4. Discussion

4.1. AI and IoT in Supply Chains

The findings of this study highlight the transformative impact of AI and IoT in optimizing supply chain management, yet their full integration remains complex due to data privacy concerns, interoperability issues, and scalability challenges. This research provides insights into the current state of IoT-AI integration in supply chains (RQ 1), the effects of IoT on AI in supply chains (RQ 2), the effects of AI on IoT in supply chains (RQ 3), the role of AI/ML in addressing digitalized supply chain challenges (RQ 4), and the AI/ML algorithms most widely employed (RQ 5). These findings not only align with previous studies but also provide critical insights into their real-world applications, challenges, and future directions for industrial deployment.

In this study, we explored the transformative potential of integrating IoT and AI in supply chains through a dual approach combining bibliometric analysis and topic modeling. Addressing five key research questions, we identified critical areas where AI-powered IoT solutions enhance supply chain efficiency, such as smart manufacturing, cold-chain logistics, and mobile edge computing. These technologies enable real-time analytics, predictive maintenance, and improved resource management, particularly in industries like pharmaceuticals, agriculture, and retail. However, challenges such as data privacy concerns, cybersecurity vulnerabilities, and interoperability issues hinder widespread adoption. Our findings emphasize the need for robust AI governance frameworks, federated learning, and global standards to address these barriers and unlock the full potential of AI-IoT integration.

Furthermore, we examined how AI and IoT mitigate digitalized supply chain challenges, including demand forecasting inefficiencies and cybersecurity risks, while also highlighting the limitations of these technologies, such as high computational costs, data inconsistencies, and model interpretability issues. We identified specific AI/ML algorithms—ranging from random forests and CNNs to genetic algorithms—that optimize various supply chain functions, including cybersecurity, demand forecasting, and predictive maintenance. Despite these advancements, the study underscores the importance of addressing scalability, energy consumption, and transparency in AI-driven IoT applications to ensure their effective deployment across industries. Overall, this research provides a comprehensive understanding of the current state, opportunities, and challenges in leveraging AI and IoT for supply chain transformation.

This study underscores the critical role of AI and IoT integration in modernizing supply chain operations across industries, while also highlighting key challenges such as cybersecurity risks, energy inefficiency, lack of AI explainability, and cost barriers that hinder widespread adoption. To fully realize the potential of AI-IoT integration, future research should prioritize the development of energy-efficient AI models for real-time IoT analytics and edge computing, alongside advancing Explainable AI frameworks to enhance transparency and regulatory compliance. Strengthening AI-driven cybersecurity measures—through anomaly detection, blockchain-enhanced security, and federated learning—is essential to protect IoT-based supply chains from cyber threats. Additionally, standardizing interoperability protocols and fostering industry–academic collaborations will be crucial for ensuring seamless integration, cross-platform compatibility, and the development of ethical AI policies. Addressing these areas will be pivotal in overcoming current limitations and unlocking the full potential of intelligent, secure, and scalable supply chains.

4.2. The Role of AI and IoT in the Digital Transformation of Supply Chains in Industry 4.0

AI and IoT, as key technologies of Industry 4.0, are driving significant changes in the supply chain. However, these changes are not limited to just these technologies; the supply chain has been influenced by various innovations both before and after their emergence. Based on our analyses, we generated Figure 13. It illustrates the historical progression and development of the supply chain over time, which is a critical factor for understanding the impact of Industry 4.0 technologies in this field. Technologies such as IoT, artificial intelligence, and blockchains have driven these changes, offering insights into the evolving landscape of modern supply chains. In the earliest stage, the integration of RFID technology and wireless sensors provided decision-makers within the supply chain with precise data concerning product attributes, such as type, quantity, customer, and location. This advancement significantly enhanced transparency and visualization across the entirety of the supply chain [123]. The application of IoT devices and sensors in supply chains resulted in the generation of huge and complex sets of data called big data, which exceeded the processing capabilities of traditional systems. Consequently, moving from early efforts to a more accelerated phase of growth, there was a surge in the utilization of big data analytics in supply chains to manage the substantial data inflow from IoT devices, as discussed in [124]. In this period, the use of Software as a Service (SaaS) providers for services like supply chain management started to become popular. This marked the beginning of the adoption of cloud computing technologies in supply chain processes, allowing organizations to leverage the flexibility and scalability of cloud services to enhance their supply chain operations. Machine learning was the next technology applied in the supply chain for predictive analytics and process optimization. By using different ML models, companies were able to learn the pattern of the buying behavior of consumers and seasonality in the data [125]. ML algorithms were used to optimize transport routes, schedules, and inventory management, leading to reduced costs and improved efficiency. Supply chains also saw the widespread integration of AI technologies for enhanced decision-making and task automation. AI algorithms are also widely applied for customer management as one of the most important elements in the supply chain [126]. Over time, the applications and usability of blockchains in supply chains increased consistently for supporting secure and transparent transactions. Blockchain technology is a decentralized ledger system that records transactions chronologically [127]. Recent advancements include the deployment of 5G networks and edge computing devices within supply chain networks to enhance connectivity and data processing capabilities. 5G is an advanced wireless network and mobile technology that offers high speed, improved reliability, enhanced capacity, complete coverage, and support for IoT devices, connectivity, and intelligent edge algorithms [128]. 5G technology has been transformative for supply chains by enabling real-time communication, improving visibility, and enhancing inventory control. Edge computing, when combined with 5G, further enhances supply chain operations by processing data closer to where it is generated. Edge computing is the name given to a set of new technologies, such as cloudlets, micro-data centers, fog, and mobile edge computing, that aim to provide storage and computational resources near to users at a network’s edge to minimize latency and response time [129]. It is anticipated that in the near future there will be a comprehensive integration of autonomous robots and vehicles into supply chain operations, marking the next phase of technological advancement within this sector.

Thus, the integration of IoT and AI has significantly influenced the development of supply chain technologies. Commencing with the adoption of RFID and wireless sensors in the early 2000s, supply chains experienced heightened transparency and improved data accuracy. The advent of big data analytics enabled organizations to efficiently handle the vast data volumes produced by IoT devices. AI applications also augmented decision-making processes and tasks, while the recent implementation of 5G networks and edge computing has bolstered connectivity and real-time data processing capabilities. Looking to the future, the projected integration of autonomous robots and vehicles by 2025 is poised to transform supply chain operations, ushering in increased efficiency and innovation.

Figure 14 also shows the evolution of policies and regulations aligned with key technological advancements, reflected across various phases of supply chain evolution. For instance, during the digital integration phase, policies like the U.S. Customs-Trade Partnership Against Terrorism (C-TPAT, 2001) and the Sarbanes-Oxley Act (2002) reinforced global trade security and data management, advocating for the digitization of supply chains. Moving to big data and cloud adoption, the Data Retention Directive (EU, 2006) and the U.S. Energy Independence and Security Act (2007) highlighted the need for data-driven operations and energy-efficient practices, enabling more localized and optimized supply chains.

As machine learning integration gained prominence, policies like the Food Safety Modernization Act (FSMA, 2011) and NIST’s Cybersecurity Framework (2014) drove the use of AI for predictive analytics, fostering transparency and security in domestic supply chains. The adoption of blockchains and enhanced security was further supported by regulations like the EU General Data Protection Regulation (GDPR, 2018) and the WTO Trade Facilitation Agreement (2017), which promoted transparency and secure data sharing through blockchain technologies.

In more recent phases, such as edge computing and 5G, initiatives like the 5G FAST Plan (2019) and the Executive Order on America’s Supply Chains (2021) underscore the governmental push toward real-time data processing and advanced IoT integration, vital for building resilient and localized supply chains. Looking forward to autonomous systems and sustainable supply chains, policies such as the European Green Deal (2020) and the U.S. Infrastructure Investment and Jobs Act (2021) focus on sustainable practices and autonomous technologies, reflecting the drive for domestication, innovation, and sustainability within national borders. The government’s efforts in developing these policies and regulations demonstrate a clear intention to further integrate Industry 4.0 technologies into supply chains.

The government’s efforts to support this technological transformation through new policies and regulations highlight the need to view this issue within a broader context. Governments once supported globalization with policies encouraging open trade and international cooperation. However, this trend has shifted significantly due to unintended consequences. Rising unemployment due to offshoring, increased data security risks, cybersecurity threats, and the disruption of global supply chains during the pandemic have pushed governments to rethink their strategies. These factors highlight the vulnerabilities of a heavy reliance on international supply chain networks, prompting a shift toward strengthening domestic capabilities. Governments invest heavily in domestic industries to reduce their dependence on foreign suppliers. The U.S. CHIPS Act, which allocates over USD 50 billion to bolster domestic semiconductor manufacturing, exemplifies this approach. Similarly, Japan offers subsidies to companies relocating production away from China, and South Korea’s Green New Deal directs funds toward local renewable energy initiatives. President Biden captured this sentiment, stating, “There is simply no reason why the blades for wind turbines can’t be built in Pittsburgh instead of Beijing”, reflecting a broader strategy to secure critical industries within the country.

In addition, environmental concerns also play a pivotal role in this shift, as governments are pushing for sustainable practices within supply chains. This environmental focus is critical as nations navigate the challenges of climate change, balancing economic growth with the need for responsible, sustainable supply chain operations. In this regard, safety standards, green regulations, and localized production are promoted to build resilient and environmentally responsible supply chains. The pandemic further exposed the fragility of global connections, showing how quickly supply chains can be disrupted. This has led to an emphasis on reshoring key industries and diversifying supply sources within national borders. By focusing on domestic capabilities, governments can better control their economic and security outcomes, protect jobs, and respond swiftly to global crises.

In conclusion, the shift from globalization to domestic capability-building reflects a fundamental change in economic strategy, driven by the need to address vulnerabilities, enhance national security, and create sustainable, resilient supply chains. As governments recalibrate policies to focus on self-reliance, they are emphasizing the integration of AI and IoT to support local innovation, secure data, and promote sustainability. This trend toward economic protectionism and technological sovereignty is reshaping the global landscape, highlighting the importance of self-sufficiency in a rapidly evolving world.

5. Conclusions, Limitations, and Future Scope

In this study, we explore the symbiotic relationship between the IoT and AI in revolutionizing supply chain operations. The integration of IoT technologies in supply chain management showcases immense promise in enhancing operational efficiency, visibility, and decision-making prowess across various sectors. However, challenges such as data quality, security issues, and integration obstacles must be resolved to facilitate widespread adoption. AI and ML algorithms have emerged as indispensable tools in overcoming these challenges, particularly in ensuring the security and dependability of IoT-driven supply chain networks. By leveraging AI for data analysis and decision-making processes, organizations can optimize logistical operations, predict disruptions, and significantly enhance productivity.

Through a meticulous research methodology encompassing bibliometric analysis and topic modeling, we scrutinize studies within the area of IoT- and AI-enabled supply chain networks from 2011 to 2024. This examination sheds light on crucial research inquiries in this domain, providing both quantitative insights into research progress and the identification of key contributors and influencers in this evolving landscape.

Nevertheless, there are certain limitations to this study that need to be considered. This study offers a broad overview of IoT and AI integration in supply chains, neglecting the exploration of specific application areas or industries. Additionally, relying solely on the Web of Science database for data collection may have introduced bias or led to pertinent studies in other reputable databases like Scopus, Google Scholar, PubMed, or Dimensions being overlooked, potentially limiting the comprehensiveness of the literature review. Furthermore, given the rapidly evolving nature of IoT- and AI-enabled supply chains, there may be new technological innovations not considered in this study.

Future research should focus on federated learning as an emerging and critical technique in AI that has gained substantial attention in recent studies related to IoT-based supply chains. The growing body of research highlights the potential of federated learning to address the inherent challenges associated with traditional machine learning, particularly in the context of IoT-based supply chains. By enabling decentralized learning across multiple IoT devices, federated learning can effectively mitigate issues of data isolation, a common barrier in distributed environments. Additionally, its ability to train models locally on devices, without requiring data to be transferred to a central server, offers a significant enhancement in terms of data security and privacy. These features make federated learning an ideal solution for advancing the integration of AI in supply chains while ensuring privacy-preserving practices. Further research should explore innovative applications, optimization techniques, and scalability in federated learning to maximize its impact on next-generation supply chain systems.

This research emphasizes the need for further comprehensive investigations to explore emerging trends and topics concerning the fusion of IoT and AI in supply chain management. By addressing critical research gaps and offering quantitative insights, this study establishes a foundation for future inquiries in this dynamic and transformative field. Ultimately, the convergence of IoT and AI holds substantial promise in reshaping supply chain management practices, fostering heightened productivity, visibility, and sustainability within the intricate global supply chains of the future.