Bibliometric Analysis of Intelligent Systems for Early Anomaly Detection in Oil and Gas Contracts: Exploring Recent Progress and Challenges

Abstract

The oil and gas industries are crucial to global economies, influencing geopolitics, driving technological advancements, employing millions, and impacting financial markets. The complexity and the volume of data generated by these industries demonstrate the need for efficient information management, where effective contract audits play a key role in ensuring market stability, transparency, fair revenue distribution, corruption mitigation, and enhancing industry integrity to attract investors. This study employs bibliometric analysis to explore the application of machine learning (ML) in detecting anomalous contracts within the oil and gas industry. This analysis identifies key research and challenges, laying the groundwork for further computational ML advancements. The PRISMA guidelines identify ML’s role from 2018 to 2023, including post-COVID-19. Principal component analysis (PCA) evaluates the bibliometric contributions of different countries and institutions. China, Indonesia, Egypt, Saudi Arabia, the University of Antwerp Operations Research Group, and the University of Pittsburgh emerge as significant contributors. These findings underscore ML’s pivotal role in fraud detection, risk mitigation, and cost savings, concluding that anomalous contract detection remains open to newer ML techniques and ongoing research.

1. Introduction

The oil and gas industries shape global economies, geopolitics, and life. These industries enable the creation of essential products, services, and companies for developing economies, meeting today’s society’s requirements [1,2]. Also, they have promoted the design of new technological advances and industrialization during the last century [1,2]. On the other hand, these industries employ millions of people worldwide and command vast investments in infrastructure, from drilling rigs to refineries and pipelines [3]. Additionally, oil exports are a critical source of revenue for many nations, influencing national budgets, the balance of trade, and exchange rates. The fluctuating prices of oil and its products can sway stock markets, impact global inflation rates, and stimulate or stymie economic growth [2,4,5]. For this reason, contracting in the oil and fuel industries plays a fundamental role in underpinning the operations and relationships that make global energy markets function efficiently [6]. Auditing state contracts in the oil industry is an essential practice that ensures transparency, accountability, and the allocation of resources [2,7]. The unique high capital investment, significant revenues, geopolitical implications, and the public trust vested in the oil sector underscores the necessity of meticulous oversight [8]. For these reasons, anomaly detection has been essential to project management and initial operation [6]. The aim of early detection is for the company to make decisions before the formal start of the project. In contracts, anomalies identify unusual activities, transactions, or patterns within procurement data that could indicate errors, fraud, or non-compliance [6,9,10]. Anomalies may be legitimate, such as honest mistakes or legal but atypical situations, or illicit, like fraud or corruption. This detection is often conducted through data analysis and machine learning techniques, looking for deviations from standard or expected patterns. It is a proactive practice to ensure the integrity and efficiency of contracting processes [11,12]. A detection system aims to pinpoint unusual behavior patterns and trends. Typically, the system produces a suspicion score, indicating the likelihood of a case being illicit. Modrušan et al. [13] emphasize the importance of preserving governmental fiscal resources, highlighting the rising adoption of sophisticated techniques to detect corrupt practices in public procurement. Also, with the auditing of state contracts, governments can ensure that revenues from the oil industry are fairly allocated and reinvested into sectors like education, health, and infrastructure [10]. However, the oil industry’s vast payments and complex supply chains can sometimes be susceptible to corrupt practices [1,2,3], so rigorous auditing can help to detect and deter such activities, upholding the industry’s reputation and ensuring fair trials. The above can attract further investment, fostering growth and development in the sector [14,15]. In this line, the challenges in the oil and gas industries, like resource depletion and transparency issues, encourage leaders and the public and private sectors to consider machine learning’s benefits [14,15].

Machine learning (ML) encompasses algorithms and models that discern patterns and make task-specific decisions using pertinent data [16]. Developing ML software entails gathering relevant datasets, picking an appropriate model, and then training this model. ML can be understood based on three major paradigms: supervised, unsupervised, and reinforcement learning [17]. This categorization, rooted in the technical literature and expert insights, showcases the flexibility of ML applications [17]. The classification and application of machine learning methods can vary between different studies [18,19]. Using machine learning for different public services has shown promising results in the health, security, and education sectors [20]. It is important to note that with the advances in information technology and the digitalization of the public procurement process (PCP), the amount of data available is increasing [13]. Advanced algorithms and associative rules can unearth potential collusive and anomalous contract behaviors. The focal point of these studies revolves around identifying corruption and irregularities in public procurement [13]. These research endeavors use data analytics techniques to categorize suspicious activities, unveiling networks and colluding entities. The effectiveness of these models largely depends on data integrity and the ability to discern pivotal warning signs. Another interesting review by Lim et al. [16] presented the importance of adopting supervised machine learning techniques to achieve high precision and confidence in identifying fraudulent transactions. However, it is essential to recalibrate and adjust these models continuously to maintain their accuracy against new dishonest tactics. These authors concluded, after a revision of credit card transactions, that there is no single perfect algorithm, i.e., each one has its pros and cons. While some algorithms can be accurate but slow in training, others might be faster but less precise. Despite the limitations and challenges, data mining and machine learning techniques are more reliable and adaptable than traditional rule-based systems.

In conclusion, auditing state contracts in the oil industry is important to ensure the sector’s ethical, efficient, and transparent functioning, playing a pivotal role in global economics and politics. Given the public and environmental stakes, such oversight is beneficial and essential for oil-rich nations’ sustainable and equitable development. Our research aims to explore the current state of machine learning (ML) adoption for detecting anomalous contracts in the oil and gas sector post-COVID-19 (2018 to 2023). The studies that utilize machine learning and address possible challenges have been reviewed. The bibliometric analysis highlights the practical implementations of such technologies and further works. So, this work contributes to an understanding of this field of research and informs different stakeholders, including policymakers, auditors, industry leaders, and technologists, about the strategic need to enhance ML integration in contracting processes to prevent fraud. It also identifies which countries or institutions are researching this field. The findings can pave the way for technological advancements and prepare the industry to integrate new technologies effectively.

This work is divided into six sections. First, the authors describe the bibliometric analysis employed based on the PRISMA flow diagram to gather documents to analyze. Then, some questions are applied to respond to the information extracted from the selected documents, and they become the study’s motivation. These questions are described below.

• RQ1: What representative works have been carried out in the oil and gas industries that help to detect anomalous contracts?
• RQ2: Which countries or institutions envisage having greater research in detecting anomalous contracts applied to the industry?
• RQ3: How have machine learning techniques contributed to detecting anomalous contracts in the oil and gas industries?
• RQ4: What approaches are applied in the literature that use machine learning and make it possible to apply in the gas and petroleum industries?

In the second section, principal component analysis (PCA) of the information extracted from the documents is performed. This analysis included a biplot representation and the dendrogram. Descriptive analysis is also performed to respond to the initial queries. In the third section, the authors describe some important works completed to detect anomalous behavior contracts in the oil and gas industries. In the fifth section, the authors comment on some of the challenges in the data, collaborations, vulnerabilities in contracts, artificial neural network (ANN) challenges, and new contract approaches that can be applied in the oil and gas industries. In the sixth and final section, conclusions and limitations are described.

2. Methodology

Bibliometric Analysis

Bibliometric analysis is a key factor for understanding and assessing the scientific output of a field using quantitative metrics. This tool enables the identification of trends, key authors, and seminal works, while also detecting collaboration networks and gaps in research. Additionally, it unveils how various disciplines are interrelated and assist in predicting future research directions. Its applications in review articles are particularly valuable, as these articles often encapsulate and reflect the state of the art in a specific area [21]. Through bibliometric analysis, one can determine which reviews have a greater impact and how research evolves over time, thereby providing a robust foundation for future research and academic decisions. Figure 1 shows a schematic diagram of the bibliometric analysis performed in this work. Firstly, based on some of the research questions (RQ) in this study, a search for articles was carried out in Scopus and Web of Science (WoS). Secondly, the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) [22] approach was selected to classify the extracted documents. Finally, a principal component analysis (PCA) [23,24] and descriptive and inferential statistics [25] were applied to conclude and propose future challenges. The PRISMA [22] flow diagram was employed to analyze the most representative documents in literature research. This methodology is important in systematic reviews, highlighting transparency and precision. It delivers a visual guide for the review process, improving reproducibility and simplifying the comparison of different review methodologies [22].

In this study, bibliometric analysis of scientific publications in the oil and gas industries related to machine learning techniques for the early detection of anomalous contracts was carried out. We used the Scopus and Web of Science (WoS) databases from 2018 to 2023. This study examined the time frame from the onset of the COVID-19 pandemic in 2019 until the post-pandemic era of 2023. Since the COVID-19 pandemic, the use of digital and smart contracts has surged across different industries, including the oil and gas sector [15]. These two databases are widely used in different literature reviews [26,27]. The search strategies were structured using key descriptors. Entry 1 was based on terms such as “machine learning” AND “contracts” AND “oil” AND “fraud detection.” Entry 2 incorporated descriptors such as “contract” AND “fraud detection” AND “machine learning algorithms” AND “oil,” AND/OR “industry” Entry 3 added a dimension with “projects”. In contrast, Entry 4 refined the focus on central terms such as “machine learning”, “contracts”, AND “fraud detection”. Using the documents provided by these databases and search engines for the academic and scientific literature, PRISMA analysis was completed [22]. Other documents were extracted and compiled from the citations of the results of Scopus and WoS [28].

The selection criteria for the selected documents for PRISMA analysis were as follows:

• Firstly, a selection of documents written in English, since most academic and industrial papers have been published in this language during the last few years [29].
• Secondly, all the documents were published from 2018 to October 2023.
• Thirdly, conference proceedings, journal articles, and working papers were included. All of these documents had to have been published in peer-reviewed journals, institutional universities, or research centers. It is important to clarify that conference papers were selected because they typically provide insights into the latest innovations and trends before they appear in journal publications.
• Fourthly, the papers’ titles, abstracts, and principal conclusions were revisited to supply the first to the third requirement. In this review, the documents had to describe strategies to detect contract anomalies using machine learning techniques in the oil, gas, or other industries.

Figure 2 shows the overall steps applied in the PRISMA diagram. As can be seen in this figure for the 211 documents gathered from the two databases, only 30 documents were used in this review using the selection criteria explained above. The systematic review of the literature showed an unencouraging panorama in the investigations focused on the detection of anomalous contracts in the oil and gas industry. It is important to clarify that works focused on cybersecurity, health, and social networks were discarded from the analysis. Our task was to examine relevant research in the petrochemical field and discuss the benefits and drawbacks of contracting, whether public or private. We reviewed several studies to be used in the analysis of contracts and fraud or to aid in this review’s argument.

Data extraction is divided into two strategies: automatic and manual extraction. Automatic extraction includes identifiers such as the DOI, study title, authors’ names, publication year, document type, and publication venue, which assist in meta-analysis and provide Supplementary Information. Manual extraction focuses on study objectives, outcomes, machine learning techniques, algorithm types, datasets used, and future research directions to address specific research questions (RQ1 to RQ4).

Overall, 53% of articles considered in this study were journal articles and reviews, 30% were conference reports, and the remaining articles were related to conference reports. Figure 3 shows the country distribution of the number of researchers who work in contract analysis (including detecting anomalies or machine learning methods used). The countries that published the most documents were Brazil, China, the USA, Russia, and Italy. Each country published more than five documents. Journal topics such as Computer Science, Business and Management, and Engineering are highlighted as the most active in researching fraud in public contracts.

Table 1. Document characteristics studied in this work.

Study	First Author (Reference)	Journal Title	Type of Document	Publication Year	Anomalous Fraud Detection in Contracts (Number of ML Methods Employed)	Do the Authors Apply ML in the Oil Industry?	Number of Citations (Auto Citations)	Number of Institutions Involved	Authors Countries	Journal Quartile a
1	Lyra [10]	Applied Network Science	Review	2022	✓ (3)	✘	2 (0)	2	2	Q1
2	Niessen [11]	2020 Seventh International Conference on Democracy & eGovernment (ICEDEG)	Conference	2020	✓ (1)	✘	8 (0)	1	1	NC
3	Chu [7]	Open Journal of Social Sciences	Review	2021	✓ (1)	✘	19 (0)	1	1	NC
4	Lima [30]	Findings of the Association for Computational Linguistics: EMNLP	Journal article	2020	✓ (2)	✘	17 (2)	1	1	NC
5	Silveira [31]	Energy Economics	Journal article	2022	✓ (5)	✓	14 (3)	6	2	Q1
6	Siciliani [32]	Information Systems	Journal article	2023	✓ (0)	✘	1 (0)	2	1	NC
7	Torres-Berru [12]	Electronics	Journal article	2021	✓ (2)	✘	2 (2)	3	2	Q2
8	Gallego [20]	International Journal of Forecasting	Journal article	2021	✓ (2)	✘	77 (5)	3	2	Q1
9	Modrušan [13]	International Journal of Advanced Computer Science and Applications	Review	2021	✓ (1)	✘	8 (0)	1	1	Q3
10	Tann [33]	arXiv	Conference	2018	✓ (1)	✘	95 (0)	2	1	NC
11	Ovsyannikova [34]	Procedia Computer Science	Journal article	2020	✓ (1)	✓	12 (1)	1	1	NC
12	Rabuzin [35]	Science and Technology Publications	Conference	2019	✓ (4)	✘	38 (3)	1	1	NC
13	Acikalin [36]	Natural Language Engineering	Journal article	2023	✓ (4)	✘	0 (0)	2	2	Q1
14	Afsharghotli [37]	Arabian Journal for Science and Engineering	Journal article	2020	✘ (2)	✓	2 (0)	1	1	Q1
15	Paltrinieri [38]	Safety science	Journal article	2019	✘ (2)	✓	228 (20)	4	4	Q1
16	Dirani [2]	Energies	Journal article	2021	✘ (0)	✓	13 (0)	1	1	Q1
17	Hendalianpour [39]	Resources Policy	Journal article	2022	✘ (0)	✓	5 (0)	3	2	Q1
18	Assad [40]	Energies	Review	2022	✘ (0)	✓	17 (0)	7	4	Q1
19	Burger [1]	Journal of Forensic and Investigative Accounting	Journal article	2022	✘ (0)	✓	2 (0)	2	1	NC
20	Sircar [5]	Petroleum Research	Review	2021	✘ (2)	✓	24 (0)	2	1	NC
21	Nai [14]	Companion Proceedings of the 23rd International Conference on Knowledge Engineering and Knowledge Management	Review	2022	✘ (5)	✘	5 (2)	1	1	NC
22	Toutounchian [4]	Safety science	Journal article	2018	✘ (0)	✓	34 (0)	1	1	Q1
23	Dahbi [41]	International Journal of Advanced Computer Science and Applications	Journal article	2023	✘ (1)	✘	0 (0)	1	1	Q3
24	Alqahtani [42]	Proceedings of the 53rd Hawaii International Conference on System Sciences	Conference	2020	✘ (0)	✘	19 (0)	2	1	NC
25	Hassan [8]	International Journal of Energy Economics and Policy	Journal article	2023	✘ (0)	✓	2 (0)	1	1	Q2
26	Shaohui [43]	IEEE International Conference on Consumer Electronics and Computer Engineering (ICCECE)	Conference	2021	✘ (2)	✘	10 (0)	4	3	NC
27	Plathottam [19]	Journal of Advanced Manufacturing and Processing	Review	2023	✘ (4)	✘	0 (0)	1	1	NC
28	Salem [17]	ACS omega	Review	2022	✘ (6)	✘	15 (4)	3	1	Q1
29	Elijah [15]	IEEE Access	Review	2021	✘ (1)	✓	33 (2)	4	1	Q1
30	Carneiro [44]	Intelligent Data Engineering and Automated Learning—IDEAL 2020	Journal article	2020	✘ (1)	✘	14 (0)	2	1	NC

^a NC and Q indicate no classification and quartiles.

shows 30 documents that were selected for this study. The table shows the journal title, type of document, whether the document studied fraud identification based on ML, the number of institutions involved, the number of authors, the countries, and the journal’s quartile (NC implies not classified). As can be seen in Table 1, only 40% of the documents listed included contract terminology or anomalous detection in the oil and gas industry, while 60% applied contract detection in other industries. The final search was performed on 31 October 2023 to gather the documents and determine the citations and auto citations. This procedure was performed manually. At the same time, the number of affiliation institutions per author (column 9), the count of the countries involved in the development of the document (column 10), and the journal quartile (column 11) belonged to the data analyses. In this work, the institution terminology referred to all universities, research centers, and industries that provided the authors with their affiliations. On the other hand, Table S1 addressed in the Supporting Information, shows an overview of the selected studies based on Table 1 that use machine learning in anomalous contract detection. In this table, the purpose of the study, the ML employed, the main findings of the research, and other comments provided by the reference are described. This table provides an overall picture of the research, which will be commented upon in the next section.

3. Results

3.1. Principal Component Analysis (PCA) Results

This work employs principal component analysis (PCA) within bibliometric analysis performed in the PRISMA selection process, demonstrating its capability to transform data volumes into key components. This technique makes bibliometric data more manageable while revealing significant trends and correlations among research topics, which is essential for informed, data-driven decision-making [23,24]. The PCA estimation was performed using STATGRAPHICS Centurion XVI. Figure 4 shows a large plot where Figure 4a indicates the country of origin and Figure 4b indicates PCA distribution components. Figure 4a shows Component 1 and 2 as the horizontal and vertical axes, respectively. The scattered points represent the projections of the data onto these components. At the same time, the arrows denote the contributions of the year, number of authors, number of institutions, total citations (without auto citations), number of documents published in a Q1 journal (called Q1 in this figure), and number of documents published in a Q2 journal (called Q2 in this figure) to these components. The number of authors and institutions appear strongly correlated and influential in Component 1. In contrast, year and Q2 significantly contribute to Component 2, and total citations show a negative influence on Component 1 and a positive influence on Component 2, indicating various underlying dynamics in the data.

On the other hand, Figure 4b presents arrows indicating the contribution and correlation of variables such as total citations, number of authors, journal article, and year with the components, revealing that journal article and year are strongly associated with Component 1. In contrast, total citations and conferences are associated with Component 2. The proximity of the points suggests clusters of observations with similar characteristics, and the perpendicular orientation of arrows like conference and year suggests independence between these variables. These analyses are linked with the dendrogram diagram, and the results are shown in Figure 5 and Figure 6.

Figure 5 presents a dendrogram illustrating the country-based PCA distribution. This dendrogram is constructed using Ward’s method (hierarchical clustering technique that minimizes the variance within each cluster) alongside the squared Euclidean distance metric (distance metric is employed to measure the distance between data points). This figure illustrates how countries are clustered according to their scientific publication profiles, considering factors such as citations, publication year, and the number of institutions involved. Proximity between groups indicates similarities in these factors, such as a comparable number of citations or similar patterns of institutional collaboration. The clustering of China, Indonesia, and Egypt suggests a similar bibliometric impact. At the same time, the later joining of Spain and Saudi Arabia on the dendrogram indicates important specialized publication profiles. Belgium and the Netherlands formed a group at a close distance, which suggests they have similar bibliometric profiles due to close collaborations between institutions or consistent citation trends over the years. It is important to note the importance of oil-rich countries such as Oman, Iran, Saudi Arabia, Egypt, and Pakistan in the bibliometric analysis. Works completed by Toutounchian et al. [4], Hassan et al., Salem et al. [8], Assad et al. [40], and Acikalin [36] demonstrate those countries’ importance to the research field.

Figure 6 shows the dendrogram of the institutions based on PCA distribution. In the dendrogram, institutions that connect at the same height indicate a high degree of similarity in their academic and research profiles, as reflected in variables such as citation numbers and publications. This is the case for pairs such as U5 and U6 (the University of Antwerp Operations Research Group and the University of Pittsburgh), which exhibit affinities in their scientific output, including close collaborations or joint publication strategies. As they move to greater heights, broader conglomerates emerge that group institutions with more diverse profiles. Within these, subgroups such as U1 to U4 form a notable characteristic that could suggest strategic alignments or similar peaks in productivity over time. Conversely, institutions that join at greater distances, like the cluster of U13, U14, U15, U19, and U20, display significant differences in their research profiles, which might be interpreted as thematic specializations or unique approaches to their publishing endeavors. These differences can be strategically significant, as institutions with high activity in top-tier journals and many citations are considered leaders in their fields, marking potential trends in research and collaboration.

A descriptive analysis is performed using the database collected in this work. This implies a graphical picture of the total number of citations related to the authors and institutions. Figure 7 shows the Q1 quartile rankings depicted by bars and the sum of citations from 2018 to October 2023. A peak in quartile rankings is seen in 2022, whereas citation counts surge to a peak in 2021 before declining. The total citation average is around 112 ± 88 (the number beside ± indicates the standard deviation). The higher standard deviation values indicate that the total citations present higher dispersion and variability in the data.

Figure 8 illustrates the tendencies of the institution and authors in contract research. Figure 8a shows the dual-axis plot comparison of the number of authors (bar plot) and the total citations (red line) across the first 10 institutions with the most citations. This figure reveals that the number of authors does not necessarily equate to citation impact, with some institutions receiving more citations despite having fewer authors. Nanyang Technological University, the Center for Corporate Sustainability (CEDON), and theNorwegian University of Science and Technology constitute the top 3 most cited institutions. On the other hand, the number of authors presents an inverse relationship with the number of citations. Universiti Teknologi Malaysia, Pandit Deendayal Petroleum University, and the University of Brasília constitute the top 3 institutions with the most authors of published scientific documents. After analyzing both variables, the standout is Pandit Deendayal Petroleum University, with disproportionately high citation numbers, suggesting that its research has significant reach or importance. The above indicates that the research impact is not solely dependent on the number of contributing authors. On the other hand, Figure 8b illustrates the proficient authors (x-axis) and the number of citations and publication year of documents (y-axis). The bars indicate the citation count for each author, while the blue solid line indicates the publication year of their works. Patrini has received the highest number of citations, at around 250, related to contract anomaly detection. There is a significant variance in citation counts among authors, with some, such as Chu, Alqhtani, Assad, and Lima, registering relatively few citations. The variation in the publication years of the authors’ works indicate that newer works are not necessarily cited more or less frequently. In conclusion, authors’ citation frequency is not directly tied to how recently their works were published. Authors with older publications can continue to accrue high citation counts, and some newer publications may have quickly gained notice.

3.2. Overview of Main Works

Figure 9 presents an overview of comments on different approaches to ML techniques. Previous research has highlighted the use of machine learning (ML) in detecting collusion, inefficiencies, and corruption, particularly in the construction industry and electronic public auctions. Lyra and colleagues [10] observed that some ML studies leverage labeled data for crime detection, using techniques like artificial neural networks, support vector machines, and random forests, with a near-even split between labeled data and exploratory analysis. Elijah and coworkers [15] have underscored the transformative potential of Industry 4.0 technologies in the oil and gas sectors, including cost reduction, production optimization, and safety and environmental impact improvements. ML techniques are particularly beneficial in contract anomaly detection, enhancing operational efficiency, safety, sustainability, and market profitability and competitiveness. Nai et al. [14] comprehensively examine ML to scrutinize fraudulent activities in public procurement, focusing on collusion indicators such as bid manipulation, overpricing, and service and material delivery fraud. The predominant ML algorithms employed are KNN, SVM, LR, RF, and GBM, with Python being the preferred programming language for these applications. These studies aim to expand fraud detection in both public and private sectors.

While ML methodologies like ANNs, SVMs, decision trees, the Isolation Forest Algorithm, and K-means show promise in the early detection of anomalous behaviors in contracts, they also bring challenges that require careful parameter tuning and consideration of the data context. The selection of an appropriate ML technique is crucial. It should align with the specific data characteristics and the objectives of anomaly detection to support decision-making processes once an oil project progresses into the execution phase [45,46].

4. Challenges in Machine Learning Techniques Applied in Contract Anomaly Detection

4.1. Data Challenges

Practical AI tools require high-quality data for training and optimal operation. Despite sophisticated algorithms, subpar data quality still needs to be addressed. While the oil and gas sectors produce vast amounts of data in different types of contracts, the quality, accuracy, and lack of labeled data still need to be addressed [5]. Human biases can infiltrate AI and ML systems, influencing data choices and the resulting interpretation. While machine learning is promising for auditing and detecting contract anomalies, it has its limitations. Despite its challenges, machine learning enhances audit speed and quality by automating redundant tasks [7]. On the other hand, many transactions, contracts, and information processes are designed for millions of clients every second, resulting in extensive databases across different systems. Instant decision-making to prevent fraud requires computationally efficient algorithms, representing a continuous challenge in all industries. Additionally, extracting features that effectively summarize historical data is vital since data can evolve. Dashboards and data analytics play a fundamental role in early decision-making.

4.2. Open Collaboration Alliances

ML algorithms emerged from an academic-practical background, fostering a culture of open sharing. While tech sectors readily adopt open innovation, the oil and gas industries remain reserved, particularly when competitors are involved. Although there is talk of promoting open source data and inter-company sharing, significant strides are yet to be made. Since universities are pivotal in AI advancements, oil and gas firms must reconsider their collaboration strategies with academic institutions [5,37].

4.3. Detection of Vulnerabilities

It is essential to involve cybersecurity to prevent and act on facing vulnerabilities in all contract procedures. Tann et al. [33] evaluated the LSTM (Long Short-Term Memory) model, an advanced variant of recurrent neural networks (RNN) to detect security threats in smart contracts, surpassing traditional symbolic analysis tools. The modeling proposed by Tann and collaborators [33] is distinguished by its weight matrices and bias vectors, allocated to specific functions such as the forget gate, input gate, and output gate, which are critical in the model’s parameter configuration. When implementing the LSTM-based machine learning technique, the authors achieved significant improvements in detecting security threats in smart contracts, surpassing traditional symbolic analysis tools. This was demonstrated by an impressive detection accuracy of 99.57% and an outstanding F1 score of 86.04%. This approach accurately identified 92.86% of previously flagged contracts as false positives. Operating at the opcode level, the LSTM model is an indispensable and exact tool for enhancing security in managing intelligent contracts. Alqahtani et al. [42] also proposed incorporating the SCSC (supply chain smart contract) to ensure integrity and compliance within the supply chain operations. The above serves as a verification and validation mechanism, conducting automatic checks to ensure that suppliers do not exceed their daily production limits and that retail points adhere to the pricing policies set by regulatory authorities. Moreover, this smart contract is an alert system detecting and notifying relevant parties of any potential fraud or breach of established production and price limits. Financially, the SCSC facilitates monetary transactions, operating as an escrow agent to ensure that payments are made according to the agreed terms, thereby enhancing the reliability and security of transactions within the supply chain [1]. Figure 10 shows a proposal diagram for the early detection of contract anomalies. As seen in this figure, automated information collection enables gathering data from multiple sources and documents related to contracts, thereby facilitating their analysis and comparison. This system compares the collected data with reference prices and statistics to detect potential overruns. Artificial intelligence and derived machine learning methods are crucial in analyzing contracts in real time and generating alerts about irregularities. Subsequently, the tax auditor assesses these alerts to verify the detected inconsistencies. This comprehensive process protects public or private resources and promotes the implementation of corrective actions to prevent the misuse of public funds.

4.4. ANN Challenges

This work shows that ANNs are typically used in machine learning modeling. ANNs are adept at processing partial or incomplete data, yet the actual impact of such missing elements on outcomes remains ambiguous. These networks can learn and establish connections between variables, although occasionally, they might resort to simple memorization, which compromises their ability to make accurate predictions. Regarding challenges, ANNs lack a universally accepted framework for their structure; this is typically developed through hands-on experience and trial and error. Moreover, the time it takes for ANNs to derive a solution is not fixed, and reaching a predefined error margin through iterations only sometimes equates to the best possible solution. However, this can be partially addressed by running the network multiple times and adjusting it to minimize the differences between its calculated outcomes and actual experimental results, thereby enhancing the performance of a specific architecture [17,47]. Karsoliya’s [48] study reveals that one to two inner layers are usually adequate for most issues, with a third layer sometimes necessary for increased accuracy. However, this adds to the network’s complexity. Also, this author proposed that the number of neurons in hidden layers should be 70% to 90% of the input layer’s independent variables, the number in hidden layers should not surpass twice that of the input layer, and their size should be between that of the input and output layers. Nevertheless, these empirical comments are not rigid rules, but rather guide the building of the ANN architecture. Different studies show that ANNs have overly complex architectures with excessive parameters and fail to analyze the ANN model’s predictive performance thoroughly [47,48]. Researchers often overlook the total number of parameters in their final models, such as bias and weights, which are essential for replicating test outcomes or applying the model to other predictive situations but are frequently underreported. This lack of transparency and detail in presenting ANN model parameters and results is a recurring issue in the literature [47]. As discussed by Faúndez et al. [47] in their work, critical issues during ANN modeling include that there is often insufficient data detail, ambiguity in the data used for training, testing, and prediction, confusion between testing and predicting phases, a lack of shared ANN code or parameter values, and incomplete information on deviations between experimental and ANN-estimated data. Also, disregard for the ratio of model parameters to training data is shown. These findings highlight the need for enhanced transparency and detail in the reporting of ANN research.

4.5. New Contracts Challenges on Oil and Gas Industry

The oil industry uses different contractual frameworks to govern exploration and extraction operations. These contractual documents outline the cooperative dynamics between sector companies and host nations, selected based on economic, risk-related, and legal considerations. Common contract types include production sharing agreements, which facilitate collaboration by dividing costs and benefits between oil companies and governments. Service agreements reimburse companies for their technical expertise and operational management, while concessions grant exclusive exploration and production rights in exchange for taking on associated costs and risks [1,5,8]. These contractual forms are tailored to comply with each locality’s regulations and market conditions. The oil and gas industries select the best contract model that aligns with the project goals and risk profiles, considering the project’s viability and profitability. Risk contracts offer greater rewards and higher risks, especially in politically volatile or geologically complex areas. In this line, there are three key stakeholders in contract awarding and oversight: the operator, the contract manager, and the contractor or project performer. The operator carefully evaluates and selects providers through inspections and feasibility studies, highlighting the need for meticulous planning and selection. Contract management emphasizes the importance of effective tendering, contract formulation, and maintenance, necessitating a capacity to adapt to changes and resolve disputes. The contractor must focus on detailed planning and cost management, including bid preparation and submission. The interaction between these stakeholders is facilitated by a contract team, stressing ongoing communication and coordination. This structure emphasizes the need for adaptable and flexible project management, highlighting the significance of collaborative efforts for project success. Thus, this structure reflects the complex interdependence of the parties involved in the process and the critical importance of effective contract management in the industry [49].

Furthermore, oil industry contracts can be categorized broadly into service and production-sharing contracts. The latter distributes the ownership and profits of the resource while leveraging the expertise and capital of international partners. Adopting these agreements, evident in countries with diverse political and economic landscapes, showcases the flexibility with which they can be adapted to each country’s unique circumstances, striving for a balance between autonomous resource management and the need to attract foreign investment. Oil-producing nations must tailor these contracts to align with income policies and economic optimization strategies [6].

Some works, as reported by Hassan et al. [8], discussed the production sharing agreement (PSA) contract requirements for the exploitation and exploration of oil. Within this contractual scheme, the government retains majority ownership of the hydrocarbons and collaborates with multinational oil entities that contribute financial resources and technical expertise to the exploration and exploitation process. Under the PSA, the international collaborating entity receives a portion of the crude extracted as remuneration for its contribution and undertaking of the business risk. In turn, the host country benefits from a segment of the extracted oil, while the international company is obliged to pay taxes on the income derived from its share of the crude. Hendalianpour et al. [39] applied the system dynamics (SD) model to assess factors impacting oil and gas development contracts with foreign investors. Factors were identified through studies, reviews, and expert feedback. After analyzing the system over five years, five scenarios were tested to determine the optimal contract type and enhancement strategies. Scenario 5, which prioritized skilled labor, advanced technology, and environmental considerations, was most beneficial for host nations and their oil entities due to its revenue advantages. However, it could have been more appealing to international investors. On the other hand, Scenario 1 faced drawbacks like human errors and lack of technology, making it least favorable. The research suggested Scenario 3, which incorporates a production-sharing contract, as a balanced choice for host countries and global oil firms, provided that the host nations select reputable contractors. Toutounchian et al. [4] emphasize the effectiveness of safety management systems (SMS) in the oil and gas industries, confirming their positive impact on reducing workplace risks and enhancing profitability. This study highlights the need for specifically defined safety budgets and improved SMS management through well-structured contract work packages. Using incentives and sanctions is recommended to ensure compliance with safety regulations and integrate safety as a criterion in the tendering process. Finally, this study suggests that neglecting safety can lead to financial losses, encouraging the rigorous monitoring of safety performance and appropriate budget allocation at all project stages.

Table 2. Comparative and improvement strategies in contract management.

Application	Industry Applied	Type of Analysis Employed	Theoretical and Practical Contributions	Disadvantages of the Study	Comments
Impact of upstream oil contracts in Iran on buy-back contracts, technology transfer and innovation [50]	Oil industry	Text analysis of 29 upstream oil contracts in Iran against a conceptual model developed from expert opinions and previous research	No significant differences among upstream oil contract types (concession, service, production sharing, buy-back) in learning and technology transfer capacity,	Presence of vague obligations and a lack of adequate performance guarantees across all contract models, including concession, service, production sharing, and buy-back contracts	The authors recommended reforming contractual frameworks to boost technological advancement and learning in the oil industry. It advocates for including specific, actionable technology and skill transfer obligations in future contracts.
Enhancing project performance and fostering sustainable development [51]	Construction industry in Hong Kong projects	Structural equation modeling (SEM) is used to study the factors influencing attitudes, collaboration intentions, and behaviors in relational contracting projects	A quantitative model to understand the impact of management commitment and relational norms on collaboration, showcasing the direct relationship between attitudes and collaborative outcomes.	The study focused on Hong Kong’s construction industry and may face limitations in broader applicability, potential survey biases, and a narrow focus on contractors, potentially missing negative collaboration outcomes.	The research highlights the importance of senior management’s role and relational norms in fostering teamwork and trust, essential for successful relational contracting and sustainable project outcomes.
Contract ambiguity’s dual role, showing that its impact on relationship performance is context-dependent [52]	Manufacturing, engineering, technical services, and trading industries, all clients of a leading law firm in China,	An empirical analysis examines contract ambiguity using survey data from 269 firms, and lawyer interviews.	The study shows that contract ambiguity impacts interfirm relationships differently depending on context: negatively in developed markets or with external uncertainties, but positively in long-term or financially significant contracts for better adaptation.	The study’s applicability is limited by its industry and regional focus and reliance on self-reported data, and potentially overlooks dynamic and qualitative aspects of contract ambiguity and interfirm relationships.	The study guides managers on the strategic use of ambiguity, based on contract and market specifics. Unaccounted for variables like communication and negotiation skills may also affect the outcomes, suggesting a need for a more comprehensive investigation.
The study investigates the effects of contractual control versus coordination on Extra-Role Adaptive Behavior (ERAB) [53]	Chinese household appliance industry	The study utilizes CFA for construct validation, PROCESS analysis to investigate fairness as a mediator between contractual governance and ERAB, and hierarchical regression to examine the moderation effect of marketization distance	The text advocates for fair contractual agreements through precise reward distribution, performance incentives, transparent communication, contribution recognition, adaptability, collaborative governance, and dispute resolution to ensure equitable, mutually beneficial relationships.	Due to differing regulatory and competitive environments and cultural norms, a single industry’s scope and cultural context might need to be better applied to other sectors or regions. The study’s design and reliance on survey data limit causal conclusions and may introduce biases, which can impact result accuracy.	The text highlights key practices for fair contracts: clear reward rules, performance incentives, open communication, contribution recognition, fairness benchmarking, collective decision-making, dispute resolution, and adaptable terms tailored to specific needs.
Identifies key factors leading to contractual disputes [54]	The document focuses on ambiguities in construction contract documents and analyzes litigation cases related to construction contracts in the Supreme Court of Canada (SCC).	The analysis encompassed 13 SCC cases, including 11 new construction projects, 1 heavy industry construction project, and 1repair and renovation work. Literature review and Supreme Court of Canada case analysis, using a “fishbone model” to structure these ambiguities hierarchically.	The study underscores the need for clear, detailed contracts and proactive measures to resolve ambiguities and improve drafting practices, aiming to minimize disputes and enhance efficiency in contract management.	The study analyzing Supreme Court of Canada construction contract cases has limitations (13 cases), including a small sample size, excluding modular/off-site construction, and needing more information on contract types and project delivery methods.	The study underscores that contract disputes often arise from unclear information, ambiguous terms, and subpar draftsmanship, stressing the importance of clear documentation and proactively addressing ambiguities.
The link between the psychological contract and job performance in Chinese higher education [55]	Analyzes the monetary dependency in the psychological contracts of a specific occupational group	Linear regression analysis quantified the psychological contract’s effect on job performance, ethical frameworks, and job performance. Statistical analyses related to the multicollinearity test, Cronbach’s alpha, KMO, and Bartlett’s tests are performed.	A strong correlation between ethics and relational terms influences perceptions of employment relationships. Ethical and emotional elements have distinct effects on job performance, with ethics mediating between psychological contracts and performance outcomes.	External factors like economic and societal changes, which could influence psychological contracts, have yet to be extensively explored. The study’s static approach may only partially represent the inherent dynamism of psychological contracts and their evolution over time.	Ethical expectations can boost engagement, satisfaction, and performance, aiding in policy development for academic staff management in higher education.
Contract preparation and execution strategies [56]	Australian construction and infrastructure industry	The study combines interviews with experts, case-specific discussions, and cross-case analysis to identify trends in collaborative contract execution. It uses the GAPPS CIFTER tool to assess project complexity, providing insights into how it affects collaborative contracting.	The research suggests that complex projects require broader, less precise contract terms to foster flexibility, while simpler ones can specify detailed terms. Early dialogue and objective alignment between clients and contractors are key contract factors. However, it notes a strategic focus mainly on project owners and contractors, with less emphasis on users and societal impact, indicating a potential oversight in maximizing stakeholder value.	Lacks extensive data on the broader stakeholder impact, including users and society, and is geographically and sectorally confined to the Australian construction industry.	Collaborative contracts that promote flexibility and joint innovation are recommended for highly complex projects to navigate uncertainties and changes effectively. The research recommends further work on how project complexity influences contract selection, the effectiveness of collaborative contracts across diverse settings, and approaches to improve engagement and value for all stakeholders.

shows comparative and improvement strategies in contract management of different types of industries [50,51,52,53,54,55,56]. This table indicates the role of clear, specific, and adaptable contracts in promoting technology transfer, collaboration, and dispute resolution across industries like oil, construction, manufacturing, and higher education. Well-designed contracts are key to boosting project performance, sustainable development, and equitable partnerships. The need for reforms to include technology transfer obligations and manage project and market complexities is also highlighted. This table explores empirical studies from Iran, Hong Kong, China, and Canada, highlighting unique contract management challenges within different sectors. Iran focuses on buyback contracts’ effectiveness for technology transfer in the oil industry. Hong Kong aims to improve construction project collaboration and performance through management and relational norms. The appliance industry faces challenges, with contractual ambiguity affecting coordination and adaptability in China. Canada concentrates on clarity in construction contracts to reduce litigation. These examples emphasize the need for contractual reform, drafting clarity, and adaptability in contract management across different contexts. This table highlights research limitations in contract studies, noting the difficulty in generalizing findings since they focus on specific sectors or regions, such as construction in Hong Kong or oil in Iran. It underscores issues with contractual ambiguity in various models, potentially leading to disputes and complicating collaboration. Additionally, it points out the risk of bias and lack of accuracy from self-reported data. The omission of important variables like communication skills, organizational culture, and industry idiosyncrasies suggests the need for further research. On the other hand, due to the growing energy demand worldwide, it is necessary to look for sustainable energy alternatives involving renewable energy. Different research works [57,58,59] explain the importance of contracting and developing new collaborations between financial and productive institutions through open innovation, essential to promote investments in environmental protection and decarbonization. The literature highlights the importance of green innovations for sustainable development, resource optimization, meeting market demands, and environmental regulations. Green financial inclusion is promoted to enhance renewable energy and support vulnerable communities, requiring effective policies and regulatory frameworks. A collective commitment at all levels is crucial to sustaining and innovating in energy sectors towards a sustainable future.

4.6. What Is Next in Early Anomaly Detection in Industry Contracts?

Further investigations will be performed on the literature exploring the impact of different approaches involving natural language processing (NLP) on predictive tasks in public procurement calls, particularly examining the performance of language-agnostic models like LaBSE (Language-agnostic BERT Sentence Embedding) and LASER (Language-Agnostic SEntence Representations) to enhance prediction accuracy. Additionally, the challenge of imbalanced data distribution across languages will be tackled through innovative data augmentation strategies, aiming to balance datasets and improve the robustness and performance of prediction models. Advancements in explainable AI will complement this effort to enhance model transparency and trustworthiness. Concurrently, research into text generation models like GPT-4 is being conducted to assist authorities in creating more effective public procurement descriptions, potentially improving procurement efficiency and transparency [36]. Finally, Python programming is the preferred language for anomalous contract detection. Due to its simplicity, freedom, and open source nature, Python stands out compared to other languages. Figure 11 provides a concise overview of key obstacles faced in anomaly detection within textual content in machine learning. These encompass the following drawn conclusions: (1) the need for models sensitive to linguistic and cultural variations due to different norms across languages and cultures; (2) the requirement for a deep contextual understanding to discern anomalies where contextual nuances might otherwise mask them accurately; (3) the scarcity of labeled datasets which hampers supervised learning in natural language processing; (4) the dynamic nature of language and the concept of ‘anomalies’ which complicate consistent identification; and (5) the complication of meaningful anomaly detection due to noise and data quality issues, the necessity of models that can adapt to intentional text manipulation by malicious actors seeking to bypass anomaly detection, and the computational strain posed by processing large volumes of text in real time which poses challenges in maintaining the balance between speed and accuracy under resource constraints [60,61,62].

5. Conclusions

The advancement of machine learning algorithms has led to the development of strategies for detecting anomalies in public and private contracts. These strategies are crucial for implementing corrective actions before project execution to prevent fraudulent activities. Specifically, in the oil and gas industries, which are pivotal globally and significantly contribute to the socioeconomic growth of countries, the methods for identifying anomalous contracts should be more extensively covered in the literature. Artificial neural networks, including recurrent approaches, are widely employed in this sector, with natural language processing methods emerging as promising tools for further work. SVMs stand out for their high precision in well-differentiated data and strong performance in high-dimensional spaces. Still, they are vulnerable to noise and become less memory-efficient with increased data volumes. KNN is straightforward to implement and valuable in non-linear contexts, though it is not the most memory-efficient option and is sensitive to outlier features and data scaling. ANNs are adaptable to linear and non-linear data and learn directly from the data. However, they require extensive training data, and their “black box” nature limits the understanding of their decision-making process. Nevertheless, this transition faces cybersecurity risks, workforce skill gaps, and environmental concerns. Despite these hurdles, the industry is progressing towards more intelligent, secure, and sustainable practices to strengthen its market standing. Machine learning in natural language processing faces significant challenges, including the need for culturally sensitive models to handle noisy and manipulated data, adapt to language changes, scale efficiently, and overcome the scarcity of labeled data. Addressing these issues requires creating advanced, adaptable, and resource-efficient algorithms.

On the other hand, the reviewed studies reveal the complexity and variety of impacts that contracts can have on inter-organizational relationships, technology transfer, project performance, and extra-role adaptive behavior. The clarity of the contract terms, the fair distribution of rewards, open communication, and the adaptability of contracts are all critical factors in fostering mutually beneficial and satisfying relationships between parties. The relevance of considering the specific context is emphasized, as contractual ambiguity and the rigidity or flexibility of terms can have different impacts depending on the market environment, industry, and cultural norms. Adopting flexible contractual clauses, prioritizing dialogue in conflict resolution, and aligning contracts with strategic goals are essential to enhance contract management in a fluid business environment. This strategy emphasizes the importance of contract management as a vital skill for organizational success and innovation, advocating for a nimble approach that fosters organizational alignment. Embracing these practices can significantly boost adaptability, collaboration, and long-term resilience and competitiveness, ensuring better project outcomes and knowledge sharing.

This study has several limitations. Firstly, the quantity of data fluctuates due to ongoing updates in the WoS and Scopus databases. Secondly, the selection of study topics occurred during the data retrieval phase, which could influence the results. Thirdly, the search terms derived from the existing scientific literature may only encompass some relevant keywords, suggesting that future research could identify new keywords. Fourthly, only articles in the English language articles analyzed during our bibliometric analysis. The above might have provided a sampling bias that could have influenced our study’s results.