Contemporary challenges and AI solutions in port operations: applying Gale–Shapley algorithm to find best matches

Abstract

Artificial intelligence (AI) developments enable human capability to deliver the same outcome at a lower cost. This research performs a high-level matching between AI solutions and challenges within the port area by developing a novel academic approach. This way, the matching is carried out more structured than when one (manager, developer, challenge owner, etc.) fulfils it based on their opinion without following any structured approach. Therefore, the study defines first a comprehensive typology of port stakeholders' challenges, which can be solved via AI solutions. This typology presents challenges, including their main issues, widespread impact, and potential solutions. A state-of-the-art review of AI solutions that can address these challenges is carried out in parallel. Secondly, this review clearly distinguishes between AI solutions based on their technology and functionality. Thirdly, this research selects an appropriate AI solution for addressing each identified challenge in the port operation by upgrading the Gale–Shapley algorithm. Finally, it shows that the most critical presented AI solutions in this study use various machine learning (ML) techniques. Besides, concerning the AI solution's reusability feature and the result of high-level matching, this research shows that the implementation phase effort can be drastically reduced by using the recently developed matching algorithm.

Introduction

Ports constitute an important economic activity in coastal areas, being gates to the world for transportation within the international trade process. They also act as a crucial connection between sea and land transport. There are many challenges in this area, making port operations relatively complex, with recent research showing that solving these challenges will bring various benefits to the local and regional economy and the environment (Jeevan et al. 2015), (DeChant 2019). Furthermore, container shipping is one of the broadest industries in the world and can reap the most considerable benefits from applying AI technologies within its operations (Chui et al. 2018). Then there would be several motivations for availing AI technology in the port and shipping industry. For instance, AI can help optimize port and shipping operations, reduce waiting times and congestion, increase vessel and cargo throughput, and improve overall efficiency (Chargui et al. 2021; Darendeli et al. 2021; Martins et al. 2020). AI technology can also enhance safety by analyzing and predicting potential hazards, preventing accidents, and mitigating risks (Lee et al. 2020; Michail et al. 2015). By automating specific tasks and optimizing operations, AI can help reduce labor, fuel consumption, and maintenance costs (Ma et al. 2020; Yan et al. 2021). AI can furthermore help reduce the environmental impact of the port and shipping industry by optimizing vessel routes, lowering emissions, and promoting more sustainable practices (Cammin et al. 2020; El Mekkaoui et al. 2020). AI technology can finally help ports and shipping companies stay competitive by improving operational efficiency and effectiveness (Niestadt et al. 2019; Shen et al. 2017).

Similarly, developers put considerable effort into AI-based solutions that can solve challenges in both logistics and ports. However, managers tend to make decisions based on their personal preferences, knowledge, or experience to implement those AI solutions or by looking at consultant advice. Therefore, it is not a reliable decision when stakeholders would implement an AI solution to solve a challenge. In addition, AI technologies are new and experimental, so there is limited information about related projects (Davenport 2018). Choosing the right AI solution requires specialized IT knowledge, which port stakeholders may lack (Murphy 2012). Sharing this valuable knowledge with port stakeholders can enhance their awareness of the appropriate types of AI solutions that can be employed to surmount challenges effectively.

A matching process is required to find the best AI solution for solving a challenge in the port area and effectively link the specific requirements of port challenges with the capabilities and features of AI solutions (Abououf et al. 2018). Nevertheless, no structured method exists to help stakeholders match AI solutions and challenges and determine the right decision. Moreover, the lack of a clear structure for matching AI solutions with existing challenges would affect the final results (DIckerson et al. 2021).

Therefore, a high-level matching under academic and scientific approach builds a structure that can assist port stakeholders in perceiving which AI solution with specific functionalities can solve which challenge(s) with certain attributes. To that purpose, a three-level approach is utilized to answer the following research question. "What is the structured method to find the best AI solution for overcoming a challenge in port and shipping industries?".

This approach contains both desk and empirical research to fill the gap in the literature. The study objective is to find the best AI solution to solve a challenge for port stakeholders. In this manner, one facet of this approach involves formulating an exhaustive list of challenges within port operations, which can be addressed through AI-based solutions. The port challenges list is compiled after conducting a comprehensive literature review. In parallel, the study generates a review of AI solutions that can address these challenges by organizing interviews. Finally, another approach's level is dedicated to developing and applying a new method for matching identified challenges and AI solutions.

This high-level matching turns into a guideline when stakeholders tend to implement an AI solution and gives insight into which AI solution from a potential list should be customized for solving their challenge. Moreover, as stated by Moscoso-López et al. (2021), all stakeholders in port areas can benefit from scientifically-founded indications in which AI technology addresses multiple problems, being the foundational future of businesses in the port ecosystem. Consequently, this can also increase the maturity level of port stakeholders from the digitalization viewpoint, including AI technology development (Sadiq et al. 2021).

On the other hand, AI developers also need access to a new academic structured methodology for decision-making regarding implementing AI solutions. Providing AI developers with this knowledge can also assist them in identifying the most promising areas for further exploration. Therefore, novel research is needed to provide advice on implementing AI solutions that address most port challenges and reduce the effort in the implementation phase by choosing the right solution right from the beginning, with the help of the matching algorithm. Accordingly, this can also contribute to raising the overall market maturity level regarding digital solutions like AI.

Besides, due to the adherence of the port and shipping industry to traditional techniques and its challenging nature for digitalization (Alop 2019; Babica et al. 2019; ESPO 2021; Fruth and Teuteberg 2017), a lack of alignment has always persisted between the expectations of stakeholders engaged in port operations and what the technology developers provide. The present study and its matching concept also have the potential to expedite this alignment process and enhance the operational efficiency of operations in ports through the availing of developed AI solutions.

This study is structured as follows: after this brief introduction, Sect. "Literature review" presents the current paper's approach, providing details regarding the steps taken to identify a matching algorithm to collect data regarding challenges and determine AI solutions' characteristics. Section 3 then presents a comprehensive literature review regarding the existing gap and the methods associated with matching algorithms. Afterward, Sect. "Maritime challenges and AI solutions" delivers the identified challenges and AI solutions in detail. Sect. "Case study: matching AI solutions and challenges" exhibits the empirical results of matching port challenges with AI solutions. Finally, Sect. "Conclusion" presents conclusions and indications for further research.

Approach

Three research steps are carried out to gather the related data for matching port challenges and AI solutions. Firstly, as this study aims to find the appropriate AI solution to address port challenges, an in-depth literature review is conducted to identify the best algorithm for matching challenges with AI solutions. The method applied to gather data for this investigation process is detailed in Sect. "Matching algorithm". The second step then collects data on a comprehensive list of challenges. The method used for this step is presented in Sect. "Identifying port challenges". Next, a list of AI solutions is also put forward in a parallel research step, which is explained in Sect. "Identifying AI solutions".

Figure 1 illustrates the approach taken by this research to match AI solutions and challenges. The dark grey-colored boxes here indicate the data collection methods followed in this study. In contrast, the light grey-colored boxes show how this data and information is collected in each intermediary step. Besides, the light grey-colored circles represent the investigation results of the collected data.

Fig. 1

Approach

The primary aim of this study is to identify the most effective AI solutions for specific challenges faced by port stakeholders. The research objective is two-fold: first, to develop a robust matching algorithm to determine the best match between the two sides (challenges and AI solutions), and second, to advise industry stakeholders on the optimal AI solutions for addressing their specific challenges. The initial objective is aligned with the research question as it aims to discover a structured approach for matching challenges with AI solutions. A second objective is to validate the designated matching algorithm. Given the dual nature of the problem, involving both AI solutions and port challenges, two distinct research steps are necessary to collect the data required for validating the matching algorithm.

The first step of the research explores the functionality of a proper matching algorithm to be used in this specific issue. As such, this step entails conducting a comprehensive literature review to identify existing matching algorithms and techniques that have successfully addressed similar problems across different domains. This step provides valuable insights into the most effective and efficient algorithms currently available.

As mentioned earlier, validating the matching algorithm between challenges and AI solutions requires equally gathering data on both sides. Therefore, as a second step, this study collects information on challenges through a desk research approach, utilizing a literature review method for two key reasons. First, a literature review provides a comprehensive and systematic approach to identifying and gathering information on various challenges. By analyzing a wide range of sources, including academic journals, books, and other publications, this study can ensure that it captures the most relevant and exhaustive information on challenges faced by port stakeholders. Second, a literature review allows for a standardized and replicable approach to data collection. This is important in ensuring that the study's findings are reliable and can be replicated in future research. By using a literature review, this study can be confident that it has collected a representative sample of challenges faced by the industry and that these challenges have been identified using a standardized approach. Besides, the academic database mainly contains applied research in which use cases from the industry have been included.

In contrast to the desk research approach used to collect information on challenges, this study employs an experimental research method to gather data on AI solutions within the third research step. There are several advantages to using interview-based research methods to collect data on AI solutions in the port and shipping industries. Firstly, interviews with developers can provide detailed and in-depth information on the functionality and technical aspects of AI solutions. This can be especially useful in identifying AI solutions' specific features and capabilities, most effectively addressing different challenges. Secondly, interviews with individual developers with direct experience or expertise with AI solutions can provide valuable first-hand perspectives on the unique use cases associated with implementing AI in the field. This can help identify practical insights and best practices that may not be readily apparent from desk research. Finally, since this paper aims to investigate innovative research on AI solutions in the port and shipping industries, interviews can be a valuable source of new and previously unidentified insights.

Matching algorithm

A thorough literature review is undertaken to discern a scientific methodology for effectively pairing challenges with AI solutions within port environments. This literature review investigates studies released since 1950. The entire literature review process is depicted in Fig. 2.

Fig. 2

Literature review process for identifying matching algorithm

Table 1 shows the results of searching "Query 1" in Fig. 2 based on journals. The right column here presents the proportion of successful results for all searches in each journal. These fractions represent the ratio of publications related to the matching algorithm to the number of publications searched for in that particular journal. For instance, in the first row of Table 1, "1/5" indicates that among five publications found in the "Discrete Applied Mathematics" journal, only one introduces an algorithm for tackling matching challenges and AI solutions. The following sub-section puts forward the method used to identify challenges that port stakeholders face within their operations.

Table 1 Overview of search results for Matching methods literature review

Identifying port challenges

Identifying and presenting port challenges begins with an exhaustive literature review (see Fig. 3). The main purpose is to provide a comprehensive and categorized list of the port and maritime transport industry's challenges, which AI can solve. There have been many studies on these terms, with this study gathering all challenges data from the port, marine, and shipping transport journals.

Fig. 3

Literature review process for identifying challenges

The resulting list of challenges is presented from a three-level perspective: micro (optimization and prediction of own operation), macro (prediction of external factors and optimization in a process that involves more than one organization), and sustainability. All these challenges are concerned with a specific port operation area, presented below:

• Waterside: This section includes infrastructure such as berth, quay, and sea carriers.

• Landside: This operation area performs activities such as importing and exporting containers, managing empty containers, and moving containers in the yard.

• Hinterland: This section is known as the truck, barge, and train operation area.

The following sub-section presents the method for identifying AI solutions implemented before, in, or beyond port areas.

Identifying AI solutions

Extra research is carried out here as a response to a lack of overview concerning whether and which AI solutions can be used in ports up to the present date. Therefore, a list of associated AI solutions has been made, which provides an overview of AI solutions that could address the challenges of the maritime transport industry, especially regarding port operations and their stakeholders.

Data regarding various AI solutions is collected through semi-structured interviews, addressing comprehensive questions. This way, knowledge from two research groups, Internet, Technology and Data Science Lab (IDlab) Antwerp and IDlab Ghent, is collected.

• The IDLab Antwerp research group is associated with the University of Antwerp and Imec. Researchers affiliated with this group perform fundamental and applied research on wireless technology, AI, and the Internet of Things (IoT).

• IDLab Ghent performs fundamental and applied research on internet technology and data science. Major research areas here are ML and data mining; semantic intelligence; distributed intelligence for IoT; cloud and big data infrastructures; multimedia processing; wireless and fixed networking; and electromagnetic, and high-speed circuits and systems.

These questions addressed to researchers have the goal of extracting the necessary information about the technology and infrastructure used to develop AI solutions, their target (optimization or prediction), the outcomes of these AI solutions after implementation, required input and data to develop them, and the future of these AI solutions. Eventually, these outcomes were gathered in a consistent report and are provided in brief in this research effort.

The AI solutions collected through experimental research within this step have already been developed to address specific challenges in port operations. The underlying concept behind adopting this particular approach is to simultaneously contribute to both practice and academia. Hence, the research gathers data pertaining to those AI solutions and broadens practical insights. It highlights additional challenges that can potentially be addressed using these solutions. Equally, existing port challenges identified during the literature review are listed within the study. This approach also enriches the existing academic literature by identifying suitable AI solutions for each of the established challenges within the literature.

Literature review

AI encompasses various branches within the field of computer science that aim to develop intelligent solutions capable of performing tasks that typically require human intelligence. Recently, organizations have moved beyond the experimental stage and are actively implementing AI technologies, leading to the widespread adoption of AI across numerous industries (Daitan 2021). Mckinsey (2019) stated that approximately 60% of companies have experienced revenue growth, while around 40% have effectively reduced costs by adopting AI technologies. However, specific barriers may cause companies to become conservative in investing further or expanding their AI capabilities. This hesitation often arises from a lack of understanding regarding which type of AI solution can effectively address the existing challenges within a company. Consequently, many enterprises have yet to embark on developing these innovative technologies.

Moreover, AI technologies are relatively novel innovations, and as projects related to AI implementation are inherently experimental, there is limited information available about these projects (Davenport 2018). Identifying the optimal AI solution to address a challenge necessitates specialized IT knowledge in AI solution development (Murphy 2012). This knowledge might not exist on the port stakeholders' side. Therefore, disseminating this knowledge among port stakeholders can heighten their awareness regarding the appropriate AI types for overcoming their challenges. Additionally, this can elevate the maturity level of port stakeholders regarding digital solution development, including AI technology (Sadiq et al. 2021). Moreover, providing AI developers with this knowledge can also assist them in identifying the most promising areas for further exploration.

Consequently, this can raise the overall market maturity level from a digitalization perspective. To that purpose, a matching algorithm is needed to effectively connect the specific requirements of port challenges with the capabilities and features of AI solutions. Besides, by considering various factors such as problem characteristics, and solution capabilities, the algorithm can identify the most suitable AI solutions for each challenge (Abououf et al. 2018).

A matching algorithm can also consider each challenge's unique characteristics and requirements and suggest AI solutions that align with those specific needs (Flach 2012). This ensures that the selected AI solutions are tailored to address the port industry's specific challenges, leading to more effective and targeted problem-solving approaches. Besides, evaluating multiple AI solutions based on their performance metrics, compatibility, and applicability to the identified challenges helps to maximize the effectiveness and impact of the AI solution in addressing the identified challenge (DIckerson et al. 2021).

In addition, by utilizing data on the characteristics of challenges and the capabilities of AI solutions, a matching algorithm facilitates data-driven decision-making. It enables decision-makers to make informed choices based on objective evaluations and comparisons, reducing biases and increasing the likelihood of successful AI implementation (Kitahara and Okumura 2021). Finally, matching AI solutions and port challenges expedites the process of identifying suitable AI solutions for port challenges, saving time and effort that would otherwise be spent on manual evaluation. It helps avoid potential trial-and-error approaches, reducing costs associated with ineffective or mismatched AI implementations (Aouad and Saritaç 2020).

Despite the existing literature on AI technologies in the port and shipping industries, which highlights the extensive research conducted in AI development, there is no framework for matching AI solutions and challenges (see Appendix 1). Only a few publications have examined the barriers to implementing AI technology in port and maritime companies. In contrast, most publications have dedicated their efforts to developing AI solutions and demonstrating the specific advantages of AI implementation in various segments of port operations and the shipping industry.

The following sub-section presents the literature review results on matching algorithms and discusses the characteristics of the best algorithm to be applied within the desired application case.

Literature review regarding matching algorithms

Since the matching problem in this study consists of two elements—AI solutions and challenges—it is expected that a bipartite matching in the graph theory will be able to address this issue. A graph in this context is made up of vertices (also called 'nodes' or 'points'), which are connected by edges (also called 'links' or 'lines') (Carlson 2020).

A bipartite graph is a graph whose vertices can be divided into two disjoint sets (Skiena 1990). There are several applications in this regard, such as matching candidates to jobs, chairs to desks, surfers to surfboards, etc.

A bipartite graph can be weighted or unweighted. In this respect, a weighted graph is one in which each branch has a numerical weight. In a weighted graph, relationships between nodes have a magnitude, which is vital for the connection. In an unweighted graph, however, the existence of a relationship is the subject (Elliot Bettilyon 2019). Subsequently, Table 2 lists the bipartite matching algorithms extracted in this literature review. These algorithms are presented as follows, with one of them selected to tackle matching challenges and AI solutions.

Table 2 Algorithms of bipartite graph

Table 2 shows three unweighted and three weighted algorithms have been identified. Nevertheless, although all the algorithms in graph theory science can solve the matching problems, convenience in implementing and being appropriate to the current study is a significant matter. Therefore, the comparison of algorithms is summarized in Table 3.

Table 3 Conclusion of literature review

This comparison was performed based on the characteristics of algorithms, namely the algorithms' time complexity, the input that needs to run algorithms, and the algorithms' limitations. Finally, one algorithm is selected for matching AI solutions and challenges.

According to the run time, the following observations are made: if there are V vertices in a bipartite graph, then n numbers belong to one set, and m numbers belong to another. This study presents vertices by V ($V = n + m$). Besides, f is the maximum flow in the graph, C is the maximum weight (cost) of edges and U is the maximum edges' capacity. U, f, and C here are positive and higher than 1. Moreover, in some instances, the runtime of algorithms has been approximately calculated. Therefore, the Gale–Shapley algorithm can probably match AI solutions and challenges in less time.

Table 3 illustrates that weighted algorithms always need more input than unweighted algorithms. Moreover, among unweighted algorithms, the Hopcroft-Karp can run with less input.

This study aims to match AI solutions and challenges based on a set of assumptions as follows: 1. Each AI solution can solve more than one challenge, with no limitation regarding the maximum number of challenges that one AI solution can solve; 2. Since finding the best match is the study's objective, it is mandatory to compare potential matching alternatives.

The Ford-Fulkerson, cycle cancelling, and Gale–Shapley algorithms satisfy assumption 1. Moreover, they can also set a capacity for each vertex to match other vertices. Therefore, if one AI solution can solve multiple challenges, assigned capacity to each AI solution can let this occur.

The second assumption refers to making the best match among all the potential alternatives. This way, if two respective AI solutions can solve a particular challenge, the algorithm must decide upon the best choices. To overcome the above matter, deciding based on each alternative's value is required. For example, weighted bipartite graphs indicate this value by the weight of each edge. Furthermore, the Gale–Shapley algorithm can define this value by each vertex's list of preferences. Accordingly, if the Gale–Shapley algorithm can match two different vertices in set A with a vertex in set B, it will match the vertex in set B with the best vertex in set A, based on the list of preferences of the vertex in set B.

According to the assumptions of the problem, Table 3 shows the algorithm that does not satisfy those assumptions. Therefore, this matter can limit utilizing this particular algorithm for matching AI solutions and challenges. In this respect, the Gale–Shapley and cycle cancelling do not have any limitations among any algorithms. Therefore, the Gale–Shapley algorithm is preferred over the cycle-canceling algorithm for several reasons.

Firstly, the Gale–Shapley algorithm guarantees a stable matching solution, meaning no incentives exist for any participant to deviate from their assigned match. On the other hand, the cycle-canceling algorithm may result in unstable solutions where participants have motives to break their matches and form new ones. Secondly, the Gale–Shapley algorithm is quicker than cycle canceling. This makes it efficient even for larger datasets. Besides, the cycle-canceling algorithm typically has exponential time complexity, making it less scalable for more significant problem instances.

Additionally, the Gale–Shapley algorithm exhibits an elegant and intuitive mechanism for matching participants based on their preferences. It optimizes the participants' preferences while ensuring stability. Conversely, the cycle-canceling algorithm may involve more complex steps and require additional optimizations to achieve similar results.

To sum up, the Gale–Shapley algorithm offers a compelling combination of stability, efficiency, and simplicity, making it a superior choice for matching AI solutions and challenges in this research.

Modified Gale–Shapley matching algorithm

This subsection modifies the selected algorithm for matching AI solutions and challenges. Accordingly, the keywords associated with the Gale–Shapley algorithm terminology are clarified as follows:

Proposing/Sending proposal: the bipartite graph has two sides, in which sending a proposal happens when one member of one side matches with the other side's members. This proposal can either be rejected or accepted by the proposal receiver.

List of preferences: each member in the Gale–Shapley algorithm must rank other members of the other side based on their preferences. This list can be incomplete.

The Gale–Shapley algorithm starts by sending a proposal from a member of one side of the bipartite graph to a member of the other side. This way, the algorithm ends when it processes the preferences list of all members of the proposer side. There are two possibilities for this: 1. The algorithm matches the member of the proposer side with one member from another side; 2. If one member of the proposer side has received rejection by its preference list members, it remains unmatched.

The input of this method is retrieved from two sides, which might conflict with each other. In this study, challenge preferences are gathered through a literature review, and AI solution preferences have been collected from developers. Hence, it is necessary to propose from one side and check whether another accepts the proposal.

The Gale–Shapley algorithm is used to solve stable matching problems. The stability of the matching between challenges and AI solutions is needed because this research tends to find the most appropriate AI solution for solving each challenge. Therefore, a pair (Challenge A, Solution B) shouldn't exist in which both members prefer each other to their partner under the Gale–Shapley algorithm.

There are two variants of the Gale–Shapley algorithm: the classical version, which solves the stable marriage problem for two sets of agents with equal size, and the "college admissions" version, which solves a related problem where a set of students are seeking to be admitted to a set of colleges (two sets of agents with unequal size) (Fenoaltea et al. 2021). The second version is considered in this study. Both variants of the Gale–Shapley algorithm are only optimal for the proposer side. For instance, if the study runs the algorithm through proposing by the challenges side, the result is optimal only for the challenges side. On the other hand, if the algorithm operates by proposing from the AI solutions side, the result is optimal for AI solutions.

Nevertheless, the equitable, stable matching problem (ESMP) aims to find a stable matching solution with avoiding bias towards either side. ESMP is more appropriate for the current study scenario than the classic SMP. Several heuristic methods have been proposed concerning ESMP. For instance, Gelain et al. (2010) developed a local search algorithm to find a fair solution for a small problem. Roth and John H. Vande Vate (1990) showed that randomly pairing two sides starting from an arbitrary matching can result in a stable solution with probability 1, but it does not guarantee fairness. Iwama et al. (2010) designed an approximation algorithm that can produce a stable solution with time complexity of ${n}^{3}+1$.

Giannakopoulos et al. (2016) created a heuristic algorithm that allows both men and women to make proposals, repeatedly leading to a fair solution. Since this study aims to find the best match for both sides, it also modifies the Gale–Shapley algorithm by running it twice (phase 1—Challenges proposed and phase 2—AI solution proposed) to avoid discrimination between the two sides of the problem. It then avails a heuristics-weighted approach to deal with inequality in each phase's results.

First, the algorithm runs by sending proposals from challenges to AI solutions at phase 1. Second, the algorithm runs in reverse, which means it operates by sending proposals from AI solutions to the challenges at phase 2. Finally, the results of these two phases might be different. Therefore, phase 3 compares the result. In other words, phase 2 validates the result of phase 1, and in case of a difference between the results of phases (phase 1 and phase 2), phase 3 will decide which phase's result is better. These phases are explained in the following sub-sections.

Phase 1. Proposing by challenges

In this phase, challenges send a proposal to AI solutions in their list of preferences. Due to assumption 1, it would be better to implement AI solutions that can solve more challenges. Hence, the algorithm in this phase sets the capacity of AI solutions as infinite, which means that each AI solution can accept the proposal of numerous challenges. Therefore, if a challenge sends a proposal to an AI solution and exists in that AI solutions preferences list, the AI solution will accept the proposal. The algorithm run in phase 1 is presented in Appendix 2.

Phase 2. Proposing by AI solutions

In phase 2, the procedure starts as in phase 1, but with two significant differences. The first point distinguishing these two phases is sending proposals by AI solutions instead of challenges, broadly affecting the algorithm. Second, during the interviews, AI solutions developers validated that each identified challenge could be solved totally by one of the identified AI solutions. Therefore, this phase considers each challenge can be paired with one AI solution. The capacity of each challenge in accepting AI solutions proposals is one.

Consequently, based on hypotheses in the Gale–Shapley algorithm, only one AI solution pairs with each challenge. Accordingly, if the number of AI solutions is less than the number of challenges, the algorithm of phase 2 must run more than once. Each time the algorithm runs here, the number of challenges that pair with AI solutions is less or equal to the number of AI solutions. Thus, the algorithm runs on several occasions to pair all the challenges. The algorithm of phase 2 is presented in Appendix 2.

Phase 3. Comparing pairs that are made in phases 1 and 2

The result of phase 1 is optimal from the challenges' perspective, and, in contrast, phase 2 provides results that satisfy the AI solutions' perspectives. Therefore, phase 3 intends to compare the results of the two previous phases while finding the most appropriate AI solution for each challenge.

First, phase 3 validates the duplicate pairs within these two phases because they are optimal for both sides simultaneously. Subsequently, this phase compares the AI solution assigned to a challenge in phases 1 and 2 in other pairs. This comparison finds the more effective pair between the two pairs with the same challenge. Although the selected pair is optimal for only one side, this is better than the other pair for the other side.

To compare pairs, the new algorithm sequence is as follows: a) the differentiation between AI solutions' ranking in the preferences list of the challenge must be calculated; b) the algorithm finds the difference of the challenge's rank in the preferences list of the AI solutions assigned to it. Since the number of challenges and AI solutions might be non-equal, to normalize the amount of (a) and (b), the count of AI solutions and challenges multiply by (a) and (b), respectively. The reason for this is that if one challenge gets the rank X among Y amount of challenges, and an AI solution gets rank X among Z amount of AI solutions, and Y > Z, then the position of the challenge is better than the AI solution's position. The following sub-section puts forward the overview of challenges AI solutions that are used further in the empirical part of this research.

Maritime challenges and AI solutions

This section presents the challenges and AI solutions identified within port areas. Table 4 shows the description of the challenge. The AI solutions implemented before in or beyond port areas by the developer in this field are listed in Table 5.

Table 4 Overview of challenges in port operation area

Table 5 Overview of AI solutions to tackle port challenges

Case study: matching AI solutions and challenges

This section puts forward a case study, along with the matching of the challenges and AI solutions identified in Sect. "Maritime challenges and AI solutions". This case study aims to find the best AI solution developed in the "COOCK smart port" project for addressing the challenges identified in the literature review of the port and shipping industries. In doing so, the designated modified algorithm takes care of the matching process in this study. "COOCK" stands for Collective Research and Development and Collective Knowledge Dissemination. Therefore, this project encourages port stakeholders to leverage AI to overcome their challenges. According to the project motto, this goal should be achieved by transferring knowledge from academia to industry. This way, the "COOCK" project plays a crucial role in advancing the AI technology perspective and enhancing the maturity level of port stakeholders in this regard.

Nevertheless, the research intends to match those AI solutions and challenges by applying the algorithm modified in Sect. "Modified Gale–Shapley matching algorithm". In other words, this study uses 30 challenges and 17 AI solutions to verify the modified algorithm. The following sub-section defines the required input to run this algorithm.

Input of matching algorithm

The necessary inputs to run the Gale–Shapley algorithm are two matrices called 'preferences lists'. The first is associated with AI solutions, while the second is provided for challenges. This way, AI solution developers rank challenges, which relies on the relevance of challenges with AI solutions. Afterward, challenges also rank AI solutions based on the relevancy of their functionality. Furthermore, the list of AI solutions and challenges preferences are presented in a table later, which presents challenges and AI solutions by codes.

AI solutions' preferences list

This research effort presents numerous AI solutions for solving challenges in port operations. However, the impact of these AI solutions on the challenges they can solve is different. For instance, one AI solution can solve challenge A better than challenge B. Moreover, it also can solve challenge B better than challenge C. Therefore, the preferences list of this AI solution is (Challenge A, Challenge B, Challenge C). In this case study, developers of AI solutions provide this preferences list about their AI solutions, with data presented in Table 6 as an AI solutions' preferences list. More information regarding the AI solutions preferences list is presented in Appendix 3.

Table 6 AI solutions' preferences list

Challenges’ preferences list

Drawn on the preferences list of AI solutions and challenges description in Table 4, the challenges' preferences list can be prepared. For instance, this section presents the reasons for ranking AI solutions for one challenge as follows:

An "Optimizing quay Crane (QC) assignment" was placed in two AI solutions' preferences lists. These AI solutions are "Resource allocation" and "Booking of slots." Both of these AI solutions rely on assignment problems but have different goals. "Resource allocation" here attempts to reduce distance, while "Booking of slots" can create faster handling time. The most significant reason for optimizing QC here is to minimize idle time and handle more containers. Therefore, it is clear that "Booking of slots" can be placed at the first slot in the preferences list of this challenge. In this way, current research prepares other challenges' preferences list as above. Finally, by utilizing these reasons, the preferences list of challenges is provided in Table 7. More information regarding the challenges preferences list is located in Appendix 3.

Table 7 Challenges' preferences list

Result of the matching algorithm

Results associated with the matching algorithm are provided in three phases based on the structure of the modified algorithm. Equally, this section also discusses the implications of these results and using the matching algorithm.

Result of phase1

According to the previous section, the challenges' preferences list was provided based on a preferences list of AI solutions. Thus, if a challenge sends a proposal to the first AI solution in its preferences list, this proposal will undoubtedly be approved. Each challenge can be matched by the first member of its preferences list in this phase. The result of running the algorithm associated with phase 1 is provided in Table 8. In this table, each challenge pairs with an AI solution except C25 and C26.

Table 8 Result of phase 1

Result of phase 2

Since the number of AI solutions is less than the number of challenges, the algorithm must run more than once. The number of challenges paired with AI solutions is less or equal to the number of AI solutions each time the algorithm runs. Thus, the algorithm runs several times to pair all the challenges with AI solutions. Therefore, the steps below present the algorithm's running process, with each step showing its result at the end.

Step 1: In this step, the algorithm of phase 2 runs just once. As a result, it matches AI solutions with challenges, except for two AI solutions. The results concerning this step and the following steps are provided in Table 9.

Table 9 Result of phase 2

Step 2: This step first removes all the challenges selected by AI solutions in the previous step. Afterward, the preferences lists must be updated, and the algorithm can run again.

Step 3: Up to this step, some AI solutions are connected with two challenges, which can be increased later in this step. As in the previous steps, if the algorithm assigns one AI solution to a challenge, the algorithm will remove the challenge from the process within the following steps.

Step 4: Only three challenges remain without an AI solution following the preceding steps. Therefore, the algorithm deletes the rest of the challenges from the process.

Step 5: Finally, the algorithm pairs the last challenge to the AI solution.

Result of phase 3

Eventually, the result associated with phase 3, which compares the result of two previous phases, is shown in Table 10. All the challenges (except C25, C26) were matched with AI solutions in this table. These pairs are the best choice for tackling challenges by AI solutions within the port operation area.

Table 10 Final result of matching algorithm

The result of the matching algorithm in phase 3 illustrates that only four AI solutions can solve more than half of the challenges in this study. The ranking of these AI solutions, due to the number of challenges that they can solve, is as follows: "Truck guidance system" (18% of challenges); "Lock optimization" (14% of challenges); "Resource allocation" (11% of challenges); and "Booking of slots" (11% of challenges) Fig. 4. These AI solutions are more critical than others in this research effort. Thus, their implementation can help port operations more than other AI solutions.

Fig. 4

Result of matching algorithms

Managerial implications

Managerial implication 1

AI solutions can be effectively classified based on their functionalities, allowing IT developers to conveniently identify which solution aligns with the specific requirements of a given challenge. However, stakeholders in the port and shipping industry may lack the necessary knowledge in this domain, hindering their ability to make informed decisions. The designed matching algorithm can be a solid foundation for organizing and structuring this knowledge. The outcome of this matching would empower anyone in the port and shipping industry, regardless of their level of expertise, to access and understand the appropriate AI solutions for addressing their unique challenges.

Managerial implication 2

Adopting AI solutions in port operations can significantly enhance productivity and managerial efficiency. With AI, the knowledge gained from developing specific models can be repurposed to tackle related challenges. Hence, it saves time and resources in data collection, model training, and computation efforts for port stakeholders. The main issue of many challenges within the current study is similar, and they are matched with the same AI solution. For instance, "optimizing ship stowage planning" and "Generating optimal yard block allocation" are linked to "Booking of slots", "Optimizing ship queuing" and "Optimizing truck queuing at the gate" are solved by solutions like "Resource allocation", "Reducing vessel waiting time" and "Reducing truck and train waiting time excess" could be tackled by "Lock optimization", "Optimizing scheduling of yard crane" and "Complex scheduling of rail-mounted gantry crane" have been matched with "Truck guidance system".

Therefore, by utilizing AI solutions to overcome one challenge, port operators can apply the same solution to tackle similar challenges with less effort, increasing overall efficiency and productivity. Moreover, those AI solutions tackling multiple challenges can potentially revolutionize the port and shipping industry. This way, implementing such AI solutions becomes cost-effective for port stakeholders and enables them to improve their operational productivity significantly. This phenomenon can be a game-changer for port operators seeking to stay ahead of the curve in the competitive global market.

Managerial implication 3

Many AI solutions presented in this study utilize ML techniques to address challenges in port operations. Among these, Reinforcement Learning (RL) is a crucial algorithm used by approximately 25% of the identified AI solutions in this research. Therefore, investing in the development of RL algorithms could bring significant benefits in the long term for port operators seeking to enhance their operational productivity and efficiency.

RL algorithms have shown promising results in optimizing various port-related processes such as resource allocation, vessel service scheduling, and container stacking. Allocating capacities to develop these algorithms further could lead to more efficient and effective port operations, saving time and resources while increasing productivity. As such, port stakeholders should consider the potential benefits of investing in RL and other cutting-edge ML algorithms.

Managerial implication 4

The challenges identified by this research belong to one of the three operation areas (waterside, landside, hinterland) or all of them. The waterside operation area possesses 12 challenges matched with 8 AI solutions. 5 challenges are associated with the landside operation area, and 3 with other AI solutions. This study identifies 7 challenges within the hinterland operation area, which have been matched with 5 distinct AI solutions.

The effort factor for solving challenges is defined based on the ratio of AI solutions to challenges (waterside: 8/12 = 67%, landside: 3/5 = 60%, hinterland: 5/7 = 71%). Therefore, the results show that less effort is needed to address challenges belonging to landside operations.

The terminal is often considered the bottleneck in the port supply chain due to its limited space. However, it is worth noting that similar AI solutions can address many challenges terminals face. Port stakeholders can leverage this insight by adopting appropriate AI solutions to overcome landside challenges, turning potential threats into opportunities. Despite the possible slowdown of port operations and increased shipping costs caused by terminal limitations, AI solutions can improve overall efficiency and productivity, helping port operators remain competitive.

Theoretical implications

Theoretical implication 1

Implementing a suitable AI solution in a port area can lead to a higher level of digitalization for port stakeholders. By matching proper AI technology to solve specific challenges, such as improving operational efficiency, enhancing security, or optimizing supply chain management, port stakeholders can better understand the potential of digital technologies and become more comfortable with using these types of technologies. This also can lead to a higher level of digital maturity and create a culture of innovation and continuous improvement. Additionally, as port stakeholders become more digitally mature, they may be more willing to invest in further digital transformation initiatives, leading to even more significant benefits for the port ecosystem as a whole.

Theoretical implication 2

There is often a gap between port stakeholders and AI developers, each with unique perspectives and objectives. Introducing AI technology to established industries can be challenging due to the inherent uncertainty of adopting novel solutions. However, demonstrating the economic viability of implementing AI in the port industry can help bridge this gap.

This research aims to identify which AI solutions with specific features can best address particular challenges in certain port operations. The matching algorithm used in this study can provide valuable input for assessing the economic feasibility of implementing AI solutions in the port industry. By analyzing each matching pair from an economic standpoint, the results can determine whether or not each match is financially viable.

Furthermore, the outcome of the investigation of economic feasibility can guide port stakeholders in identifying which AI solutions with specific attributes can be most beneficial for their particular type of operation. This information can inform decision-making processes and help port actors to make more informed choices about implementing AI solutions in their operations. By leveraging these findings, port stakeholders can become more knowledgeable about the potential benefits of AI technology and make strategic decisions to drive economic growth and efficiency in their operations.

Conclusion

Contemporary port stakeholders face various difficulties within their operations. In contrast, technology providers (recently focusing on AI, big data, or ML) claim to solve numerous port issues. In this regard, there is a lack of a marketplace in which the supply can meet the demand, as in traditional markets. Moreover, there are difficulties in structurally matching the right AI solution provider with the respective challenge owner. This research develops and applies a new academic approach that structurally identifies the AI solution that solves the appropriate challenge within port operations.

This research sets pioneering steps in matching AI solutions with challenge owners. It has the following tangible results: first, it conducts an in-depth literature review to investigate matching algorithms. Accordingly, the research intends to identify innovative methods to link the two market sides (challenges and AI solutions), in which the Gale–Shapley algorithm is found to provide the best results. Since this algorithm is only optimal if applied from the perspective of one side of the market (the proposer's side), this research contributes to the literature associated with this algorithm by using these principles, developing a novel integrated sequence that runs this algorithm from two perspectives, then provides comprehensive matching results satisfying the conditions of the two market sides.

Secondly, this research carries out some desk/empirical research to provide an overview of contemporary challenges and AI solutions within ports. It briefly describes the challenges by presenting their underlying issues, impact, and potential solution characteristics. Similarly, an overview of AI solutions is exhibited. In this respect, an intermediary observation is made that while AI technology is highly likely to be applied in several port operations areas, technology providers integrate their conceptual models in highly focused and dedicated AI applications. For instance, the technological concept that uses AI to "predict vessels' ETA" is almost equally helpful in predicting the delay of trucks and trains in hinterland operations.

As its main and third outcome, this research identifies the AI solutions matching contemporary challenges within port operations by applying the newly developed sequence based on the Gale–Shapley algorithm. Accordingly, the data regarding which AI solutions can solve these port challenges is collected due to face-to-face meetings between technology providers and challenge owners. Subsequently, the in-depth literature review collects data regarding the characteristics and preferences of challenge owners.

The results here show that the concepts of several AI solutions also tackle challenges from other port operational areas than initially intended. For example, a "Truck guidance system", developed to reduce queues in both landside and hinterland operational areas, solves most port challenges (5 out of 30). Moreover, a "Lock optimization" solution using AI and operational in the waterside area of ports tackles 4 out of 30 port challenges in total. Similarly, the use of a solution that enables digital "booking of time slots" at terminals and/or "Recourse allocation" can tackle an equal share of challenges (3 out of 30). Besides, the concept of "Booking of time slots" is matched with challenges that solve issues in container warehousing, with "Resource allocation" linked to queuing problems. On the other hand, when selecting one port operational area, the landside operation area would benefit the most from implementing an AI solution.

Furthermore, the foundation of these critical AI solutions is implemented once. Concerning the reusability feature of AI technologies, stakeholders can customize those AI solutions with less effort to solve the challenges. Moreover, the AI solution overview shows that the key technology these solutions use is ML, specifically the RL algorithm, which can be worth investing in.

This study develops a structured methodology to match port challenges and AI solutions. The main limitation of applying and gaining meaningful results from this method lies in the data collection. Although the buildup of the preference list from both challenges and AI solutions perspectives is supported by strong arguments, there is no theoretical paradigm for collecting this data. Therefore, as a future step, new research is undertaken to employ a scientific method that objectively obtains the preferences of each market side. This method will consider the characteristics of solutions and challenges for defining these preferences. These measures can then make the input of this research more reliable, and the final result can be more accurate.

Appendices

Appendices

Appendix 1

See Table 11.

Table 11 Overview of the literature adhering to the AI technologies in port and shipping industries

Appendix 2

Algorithm phase 1

Proposing by challenges

See Fig. 5.

Fig. 5

Algorithm phase 1 flowchart

Algorithm phase 2

Proposing by AI solutions

See Fig. 6.

Fig. 6

Algorithm phase 2 flowchart

Algorithm phase 3

Comparing pairs that are made in phases 1 and 2

See Fig. 7

Fig. 7

Algorithm phase 3 flowchart

Appendix 3

See Tables 12 and 13.

Table 12 AI solutions preferences list details

Table 13 Challenges preferences list details

Appendix 4

Table 14 presents the process of finalizing pairs due to the algorithm phase 3.

Table 14 Comparing process