IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 14, No. 2, April 2025, pp. 1663~1672

ISSN: 2252-8938, DOI: 10.11591/ijai.v14.i2.pp1663-1672  1663

Graph-based methods for transaction databases: a comparative

study

Wael Ahmad AlZoubi1, Ibrahim Mahmoud Alturani1, Roba Mahmoud Ali Aloglah2
1
Department of Applied Sciences, Ajloun University College, Al-Balqa Applied University, Ajloun, Jordan
2
Department of Management Information Science, Amman College for Financial and Managerial Sciences, Al-Balqa Applied University,
Amman, Jordan

Article Info ABSTRACT

Article history: There has been an increased demand for structured data mining. Graphs are
among the most extensively researched data structures in discrete
Received Jul 29, 2024 mathematics and computer science. Thus, it should come as no surprise that
Revised Nov 3, 2024 graph-based data mining has gained popularity in recent years. Graph-based
Accepted Nov 14, 2024 methods for a transaction database are necessary to transform all the
information into a graph form to conveniently extract more valuable
information to improve the decision-making process. Graph-based data
Keywords: mining can reveal and measure process insights in a detailed structural
comparison strategy that is ready for further analysis without the loss of
Data mining significant details. This paper analyzes the similarities and differences
Graph among four of the most popular graph-based methods that is applied to mine
Rule mining rules from transaction databases by abstracting them out as a concrete high-
Structured data level interface and connecting them into a common space.
Transaction database
This is an open access article under the CC BY-SA license.

Corresponding Author:
Wael Ahmad AlZoubi
Department of Applied Sciences, Ajloun University College, Al-Balqa Applied University
Ajloun 26816, Jordan
Email: [email protected]

1. INTRODUCTION
Graph-based methods for a transaction database are necessary to transform all the information into a
graph form to conveniently extract more valuable information [1]–[3]. Graph-based data mining can reveal
and measure process insights in a detailed structural comparison strategy that is ready for further analysis
without the loss of significant details [4]. In addition, the graph-based methods process can be considered as
a process mining method.
This research aims to systematically understand the trade-offs among graph-based methods for mining
transaction datasets by comparing them. There are four main methods to mine transaction datasets using graphs,
they are: clique percolation system [5], adjacency matrix [6], graph neural network (GNN) [7] and network-
based visualization [8]. Each one of these methods follow the same general idea: constructing a graph that
captures the relations between different parts of the structured data. Despite the diversity of methods and the
variations in the exact form that the final task-related graph takes, some clear organizing principles emerge.
A transaction database is a collection of records; each record contains pieces of data. These records
are also called transactions. A graph database is a database management system that uses graph structures to
store, map and query relationships. Every element contains a direct pointer to its adjacent element and can
also be used to perform search in constant time using hash index [9]. The transaction database management
system supports transactions from multiple customers and does not contain any customer master data. A
transaction database does not allow for the full capabilities of a transaction to be represented. It abstracts the

Journal homepage: http://ijai.iaescore.com

1664  ISSN: 2252-8938

transactions to a form that is compatible with the machinery of the transaction database. A graph database
attempts to capture the full detail of a transaction [10].
We outlined a comparative study on the graph-based approaches for mining different useful patterns
by growing algorithms in case of the transaction database [11]. Table 1 briefly explains some of the main
characteristics of these methods. This table helps to focus the different features and applications of each
method for network analysis and visualization.

Table 1. Graph-based mining methods' characteristics

Method Description Uses Graph representation Interactivity
1. Clique System used to find and Identifying Focuses on identifying Minimal interaction:
percolation analyze complete sub graphs interconnected groups cliques, not a direct manually inspecting
system (cliques) in networks, focusing and communities visual representation. identified cliques is
on identifying fully connected within networks. frequently necessary.
groups of nodes.
2. Adjacency This method represents the Studying network Represents Static representation,
matrix relationships among the nodes construction accurately, connections between needs manual
in 2D array (matrix) showing computing network nodes in a matrix adjustment for
connections as binary values metrics like degrees form. network changes.
(presence or absence of edges). and shortest routes.
3. GNN method Neural network approach to Node classification, Learns node and edge Interactive for network
learn node and edge features link prediction, and features using deep exploration and
for prediction and classification community detection learning techniques. predictive tasks.
tasks in networks. in complex networks.
4. Network-based Visual representation technique Visual exploration of Provides visual Highly interactive,
visualization for networks, showing nodes network structures, insights into network allows real-time
and links in a graphical and understanding topology and exploration and
interactive manner. relationships and dynamics. analysis.
identifying patterns.

This study covers graph-based algorithms for data analysis of transaction databases and provides a
comparative analysis regarding selected property descriptors. Retail datasets of 1000 transactions will be
taken as a case study to clarify the role of each method in extracting the desired association rules, compare
among them and so enhance the decision-making process. To the best of our knowledge, we introduce a
comparative study of the graph-based methods used to discover rules from transaction datasets.
The overall structure of the research is organized as follows. Section 2 talks about the main graph-
based methods for transaction datasets. Sction 3 explains briefly the research methodology. Section 4 discusses
the comparative analysis of these methods. Section 5 the results of previous studies were comprehensively
reviewed and analyzed using the criteria described there. Lastly, section 6 concludes this paper.

2. GRAPH BASED METHODS FOR TRANSACTION DATASETS

As we mentioned earlier in the introduction, a dataset of retail sales will be studied and analyzed since
this type of datasets has been developed safely with the coming of president data science methods and tools
[12]. Nowadays, retail enterprises create advanced techniques to derive meaningful conclusions from massive
volumes of transactional data [13]. The most common among these techniques are: the clique percolation
system, adjacency matrix analysis, GNNs, and network-based visualization. These algorithms offer powerful
ways to uncover hidden patterns, complex relationships between products and customers will be discovered,
and totally improve decision-making. We will examine how these techniques can be successfully used in retail
sales environments to enhance consumer engagement, optimize strategies, and spur business growth. Retail
companies can improve customer satisfaction, boost operational efficiency, and improve their marketing
strategy by incorporating these tactics and analyzing the links and trends in their sales data. In the following
sub-sections, we will describe briefly how these techniques are used in the context of retail sales dataset.

2.1. Clique percolation method

The clique percolation method is a common method for examining the overlying public construction
of networks. The clique percolation system can be used in retail sales to find products or category clusters
that are commonly purchased together, as well as significant correlations between them. For instance, it can
reveal product groups that are frequently purchased together or close connections between categories.

2.2. Adjacency matrix

The adjacency matrix offers a matrix representation of nodes and their pairwise relationships based on
transaction interactions showing connections as binary values (existence or nonexistence of edges). In retail

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1663-1672

Int J Artif Intell ISSN: 2252-8938  1665

sales data, links between items or product categories are represented by the adjacency matrix. A product or
category is represented by each row and column, and the matrix shows whether there is a relationship between
them or not. You can use this matrix to look at relationships and find fresh patterns in sales data.

2.3. Graph neural network

In discovery of complex associations from transaction data, the GNNs play an important role in
finding hidden rules that represent the relations among products. GNNs signify the transactions as graphs to
forecast conclusions such as customer comportment, product commendations, or deceitful activity. GNN
algorithms are used to assess retail sales data and anticipate buyer behavior by means of product relationships
and prior purchase patterns. GNNs are useful for understanding complicated linkages between goods and
consumers as well as examining how marketing and promotions affect these connections.

2.4. Network-based visualization

This method gives graphical depiction for networks, displaying nodes and edges in a graphical and
collaborative way. Visual representation and analysis of the outcomes of the GNN, adjacency matrix, and
clique percolation system predictions in retail sales data are done by network-based visualization. It helps
analysts and managers make based on data strategic decisions by offering an illustration of the complex
relationships among products.

3. RESEARCH METHODOLOGY
The same set of data across all tested methods is used during the comparative study. This approach
ensures fairness and consistency in evaluating the performance of different graph-based methods for mining
transaction datasets [14]. The main graph-based methods to mine rules from transaction datasets, i.e., clique
percolation, adjacency matrix, GNN and graph visualization are tested over the same set of transactions. An
intuitive choice is to use a graph database as a new type of database and thus this technology has generated
great attention. There are several surveys in the literature that summarize the existing graph databases and
their applications [15].
A comparative study focusing on graph-based methods used for mining transaction datasets involves
evaluating various techniques within this domain will be discussed. Figure 1 highlights the main steps to
discover the find out the best choice by do an efficient comparison among graph-based methods from
customer data. These steps improve the accuracy and truth of the comparative study's results, this will lead to
worthy remarks into the best method(s) for extracting desired rules from transaction datasets. The following
subsections talks briefly about each one of these steps.

Start Choose the dataset

Clean the dataset

Dataset analysis
No

Is dataset
uniform?

Yes
Apply methods

Do comparison

Results End

Figure 1. The flowchart of the experimental methods applied

3.1. Dataset selection

Choosing the right data set is not as simple as many people think, as there are criteria for choosing
the appropriate data set, such as being compatible with the field of interest or study, and it must afford

Graph-based methods for transaction databases: a comparative study (Wael Ahmad AlZoubi)
1666  ISSN: 2252-8938

adequate transactional data. The chosen dataset should also be complete, accurate and free of outliers. The
same set of data will be used for each method under investigation during the comparison analysis. This
methodology guarantees impartiality and uniformity while assessing the efficacy of various graph-based
techniques for transaction dataset mining.

3.2. Dataset cleaning and preprocessing

Data cleaning is an important step in improving the superiority of the data and confirm that we can
infer eloquent rules. To guarantee consistency and quality of data, clean up and preprocess the dataset.
Depending on the requirements of each approach, this stage may involve resolving missing values,
normalizing data, and encoding categorical variables.

3.3. Apply methods on uniform dataset

When the selected dataset is ready to be used, i.e. it is cleaned from any outliers or missing values,
the graph-based methods will be used directly to assist in making right decisions and the overall mining
process will be improved. Utilize the standardized dataset with every graph-based technique, following the
same guidelines. To ensure comparability and remove bias, all methods must use the same preprocessing
procedures and settings.

3.4. Analysis and evaluation

It is very important to analyze and evaluate the results after applying the different graph-based
methods on the selected transaction dataset. This phase aids us realize the efficiency of the chosen approach,
measure the performance of each method, and find what must be improved. Gather and examine each
method's output according to predetermined assessment criteria. These criteria might include outcomes
interpretability, computational efficiency, scalability in managing big datasets, and accuracy of transaction
pattern recognition.

3.5. Comparison
The performance of the chosen graph-based methods must be compared depending on five criteria,
they are: scalability, accuracy, complexity, interpretability and versatility to be able to determine which one
is the best in dealing with transaction dataset. Based on the evaluation metrics, compare how well each
technique performs. Determine the advantages and disadvantages of each approach in comparison to the
others, emphasizing any compromises that might affect how well-suited each is for a given kind of
transactional data analysis.

3.6. Comparative analysis of graph-based methods

Graph-based methods have been used extensively with transaction databases. For this comparative
study, we focus on the most widely used close n-vertices adjacency graph representation. This representation
defines a graph where each node represents an item in the database and n-vertices are qualified as adjacent to
each other if they appear together in a transaction. It is also referred to as the unique-itemset-content-
compatible graph (UCC graph) [16], [17].
Retail dataset is one of the popular datasets used in data analysis and pattern mining studies in retail
and sales. This group includes data on purchases that are typically recorded through point-of-sale (POS)
systems in stores and shops. Data usually includes:
− Product information: such as name, description, and category.
− Customer information: such as age, gender, and location of residence.
− Purchase details: such as date, time, and amount paid.
− Store information: such as location, branches, and departments.
− Payment methods: such as cash, credit cards, and electronic payment.
Using a retail dataset can help analyze customer purchasing behaviors, discover common patterns in
purchasing, forecast product demand, and improve inventory management and marketing strategies. This kit
is ideal for research studies and business analysis in the retail industry [16]–[18]. It will be efficient to assess
and select the best graph-based technique for generating rules from transactional datasets by applying this
structured comparative study, considering the features of the dataset and the users' unique requirements.
Table 2 is an expanded table that includes the evaluation for each method: clique percolation system,
adjacency matrix, network-based visualization, and GNN. This table provides a comprehensive overview of
how each method is evaluated in terms of analysis, visualization, and prediction capabilities based on the
available data.

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1663-1672

Int J Artif Intell ISSN: 2252-8938  1667

Table 2. The evaluation of the graph-based mining methods from transaction datasets
Method Evaluation Details of evaluation
1. Clique percolation Analysis of discovered cliques and Evaluation of clique size and frequency comparison across
system comparison against expectations and various clique percolation system settings (e.g., changing k
requirements if applicable).
Effectiveness of cliques in predicting future network or
data behavior.
2. Adjacency matrix Analysis of relationships between Analysis of existing relationships in the adjacency matrix.
categories and measuring relationship Measurement of relationship strengths between categories
strengths based on values in the matrix. Comparison of adjacency
matrices under different bases (e.g., quantity or price).
3. Network-based Visual understanding of relationships Visual understanding of relationships between different
visualization and representation of developments over categories. Representation of developments over time if
time using temporal network visualization. Comparison of
different network visualizations based on drawing
techniques and emphasizing key relationships between
categories.
4. GNN Improvement in product categorization Evaluation of GNN's ability to control network data for
or sales prediction based on networks improving product categorization or sales prediction.
Examination of GNN's performance in learning intricate
relationships between categories based on available data.
Comparison of GNN results with traditional methods.

4. RESULTS AND DISCUSSION

In this section, it is explained the results of research and at the same time is given the
comprehensive discussion. Results can be presented in figures, graphs, tables and others that make the reader
understand easily [19], [20]. In the literature [21]–[25], there are many studies about the different graph
based methods for transaction datasets, we used the same set of data for each method under investigation
during the comparison analysis. This methodology guarantees impartiality and uniformity while assessing the
efficacy of various graph-based techniques for transaction dataset mining.
Five different criteria were used to offer a complete structure for allocating numbers to the tables
that reflects an exhaustive evaluation of the effectiveness of each technique in relation to network data
analysis and visualization [4], [26]. The criteria are:
− Scalability: assesses how well each technique can manage increasing amounts of data without
sacrificing efficiency and concert.
− Complexity: evaluates each method's computational cost and resource usage (memory and CPU time).
− Accuracy: evaluates each method's capacity to produce accurate and dependable outcomes in tasks
involving investigation and presentation.
− Interpretability: evaluates the ease of comprehension and interpretation of the outputs and outcomes
produced by each method.
− Versatility: examines the adaptability of each method to a broad range of activities and applications.
Each of these criteria will be tested separately for each of these methods and then the results will be
compared as in the following sections.

4.1. Scalability
Each method's scalability differs greatly depending on how it is designed and intended to be used.
The modest scalability of the clique percolation system makes it appropriate for medium-sized networks, but
it might be problematic for very large datasets [26], [27]. The adjacency matrix, on the other hand, shows
good scalability and is effective for big, static networks, but it could need a lot of assets for networks that are
dynamic [27]. When properly designed, the GNN exhibits significant scalability as well, making it a viable
option for efficiently processing huge datasets [28], [29]. Depending on the amount of the dataset and the
display capabilities, network-based visualization [30] provides strong scalability for visual exploration,
making it easier for users to explore network structures easily. These findings aid in the suitable technique
choosing, considering the scalability requirements for analysis or visualization chores.
Based on the allocated numerical values, this representation makes it easier for consumers or
researchers to understand how the procedures differ from one another in a more structured way. It makes
decision-making easier depending on certain analysis requirements or intended results. Figure 2 and Table 3
illustrate graphically the scalability of each one of these methods on the selected retail dataset.

4.2. Complexity
The complexity degree of each method is shown by the "complexity" results. The clique percolation
system exhibits low complexity by using simple methods that are effective in terms of processing speed and

Graph-based methods for transaction databases: a comparative study (Wael Ahmad AlZoubi)
1668  ISSN: 2252-8938

memory utilization. The complexity of the adjacency matrix ranges from low to reasonable, depending on the
extent of the entire network and memory needs [31]. Because they employ deep learning techniques, GNNs
exhibit enormous complexity, requiring substantial processing resources and a lengthy training period [7], [32].
Network-based visualization is low to moderately complicated, with simple display operations at the base [33].
Large networks or interactive functionality may call for additional resources. The findings shed light on how each
technique manages the complexity and processing demands of network data analysis and visualization. Figure 3
and Table 4 illustrate graphically the complexity of each one of these methods on the selected retail dataset.

Scalability
5
Scalability Level

4
3
2
1
0
Clique Percolation System Adjacency Matrix Graph Neural Network (GNN) Network-based Visualization

Graph based method

Figure 2. Graphical representation of the scalability among the graph-based methods for retail dataset

Complexity
5
Complexity Level

4
3
2
1
0
Clique Percolation System Adjacency Matrix Graph Neural Network (GNN) Network-based Visualization

Graph based Method

Figure 3. Graphical representation of the complexity among the graph-based methods for retail dataset

Table 3. Scalability of graph-based methods Table 4. Complexity of graph-based methods

Method Scalability Method Complexity
Clique percolation system 3 Clique percolation system 1
Adjacency matrix 4 Adjacency matrix 2
GNN 4 GNN 5
Network-based visualization 3 Network-based visualization 2
Explanation of values: Explanation of values:
Scalability: Complexity
1: Low scalability 1: Low complexity
2: Moderate scalability 2: Low to moderate complexity
3: High scalability 3: Moderate complexity
4: Scalable for large datasets 4: High complexity due to deep learning techniques
5: Highly scalable with appropriate architecture 5: Very high complexity

4.3. Accuracy
The "accuracy" results show how accurate each method is. The clique percolation system is a good
tool for recognizing communities within networks since it shows good accuracy in identifying cohesive
groups, or cliques. The adjacency matrix is a visual aid that makes node connections easier to understand
while offering excellent accuracy in computing network metrics like node degrees and shortest paths [27].
When learning node and edge features, GNNs demonstrate exceptional accuracy, which makes them useful
for intricate pattern recognition applications [7], [29]–[31]. Depending on the methods used and the level of
user experience, network-based visualization exhibits medium to high accuracy in displaying network
architecture and spotting patterns [33]. These points demonstrate how each technique complies with
requirements for accuracy while examining and displaying network data. Figure 4 and Table 5 illustrate
graphically the complexity of each one of these methods on the selected retail dataset.

4.4. Interpretability
The term "interpretability" describes how simple and intuitive it is to understand and examine the
outcomes of any given method [4], [26]. Because the clique percolation system mainly finds cohesive groups

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1663-1672

Int J Artif Intell ISSN: 2252-8938  1669

(cliques) without offering a clear visual representation, it is difficult to intuitively grasp the results, which
contributes to its low interpretability [27]. The adjacency matrix, on the other hand, provides excellent
interpretability by graphically depicting node connections, making it possible to comprehend network
interconnections and structure with clarity [28]. Given that they learn intricate node and edge properties,
which may call for more in-depth research to properly interpret, GNNs exhibit intermediate interpretability
[7], [29]–[34]. High interpretability is achieved using network-based visualization, which makes it simple to
identify important network properties by providing a clear visual understanding of network topology and
patterns [35]. These variations highlight how the interpretability of each approach meets various
requirements for efficiently understanding and analyzing network data. Figure 5 and Table 6 illustrate
graphically the interpretability of each one of these methods on the selected retail dataset.

Accuracy
5
Accuracy Level

4
3
2
1
0
Clique Percolation System Adjacency Matrix Graph Neural Network (GNN) Network-based Visualization
Graph based Method

Figure 4. Graphical representation of the accuracy among the graph-based methods for retail dataset

Interpretability

5
Interpretablity Level

4
3
2
1
0
Clique Percolation System Adjacency Matrix Graph Neural Network (GNN) Network-based Visualization
Graph based Method

Figure 5. Graphical representation of the interpretability among the graph-based methods for retail dataset

Table 5. Accuracy of graph-based methods Table 6. Interpretability of graph-based methods

Method Accuracy Method Interpretability
Clique percolation system 4 Clique percolation system 2
Adjacency matrix 5 Adjacency matrix 4
GNN 5 GNN 3
Network-based visualization 4 Network-based visualization 5
Explanation of values: Explanation of values:
Accuracy: Interpretability:
1: Low accuracy 1: Low interpretability
2: Low to medium accuracy 2: Moderate interpretability
3: Medium accuracy 3: High interpretability
4: High accuracy 4: High interpretability; matrix format visually represents node connections
5: Very high accuracy 5: Highly interpretable; provides basic visual insights

4.5. Versatility
The degree to which a method can be tailored to a variety of activities and applications is referred to as
its versatility. With its narrow scope of applicability, the clique percolation system is mainly useful for studying
organized groups in networks. For a variety of analytical and mathematical activities requiring the structural
representation of the network and the computation of different metrics, the adjacency matrix provides good
adaptability [36]. GNNs are very versatile; they can handle a wide range of jobs because they can recognize
intricate patterns and adjust to various kinds of network input [37], [38]. Additionally, network-based
visualization offers great variety by enabling interactive and visual network exploration and analysis, which
makes it easier to fully comprehend network patterns and structures [39]. These differences show how each

Graph-based methods for transaction databases: a comparative study (Wael Ahmad AlZoubi)
1670  ISSN: 2252-8938

approach fits requirements for network data analysis and visualization in various application contexts. Figure 6
and Table 7 illustrate graphically the versatility or adaptability of each one of these methods on the selected
retail dataset. The retail dataset used in the literature contains 1000 transactions distributed over three main
categories [25], i.e. clothes, electronics and cosmetics or beauty tools. Table 8 shows some data from the retail
dataset chosen in the experiments. The schema or the description of the dataset is given in Table 9.

Versatility
6
Versatility Level

0
Clique Percolation System Adjacency Matrix Graph Neural Network (GNN) Network-based Visualization
Graph based Method

Figure 6. Graphical representation of the versatility among the graph-based methods for retail dataset

Table 7. Versatility of graph-based methods

Method Versatility
Clique percolation system 1
Adjacency matrix 3
GNN 4
Network-based visualization 5
Explanation of values:
Versatility:
1: Limited versatility
2: Moderate versatility
3: Versatile for various tasks
4: Versatile for various tasks including node classification and link prediction detection
5: Highly versatile for exploratory analysis

Table 8. Retail dataset used in the comparison

# Transaction ID Date Customer ID Gender Age Product category
0 1 2023-11-24 CUST001 Male 34 Beauty
1 2 2023-02-27 CUST002 Female 26 Clothing
2 3 2023-01-13 CUST003 Male 50 Electronics
3 4 2023-05-21 CUST004 Male 37 Clothing
4 5 2023-05-06 CUST005 Male 30 Beauty
Quantity Price per unit ($) Total amount
0 3 50 150
1 2 500 1,000
2 1 30 30
3 1 500 500
4 2 50 100

Table 9. Retail dataset schema

# Attribute Count Null Data type
0 Transaction ID 1,000 non-null Int64
1 Date 1,000 non-null object
2 Customer ID 1,000 non-null object
3 Gender 1,000 non-null object
4 Age 1,000 non-null Int64
5 Product category 1,000 non-null object
6 Quantity 1,000 non-null Int64
7 Price per unit 1,000 non-null Int64
8 Total amount 1,000 non-null Int64

5. CONCLUSION
Since the development of sophisticated data science methods and tools, retail sales analytics has
undergone substantial change. Retail businesses now have access to advanced techniques for deriving useful
conclusions from massive volumes of transactional data. The clique percolation system, adjacency matrix

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1663-1672

Int J Artif Intell ISSN: 2252-8938  1671

analysis, GNNs, and network-based visualization are important methods among these. These approaches
provide effective means of revealing latent patterns, comprehending intricate interactions between goods and
consumers, and eventually improving decision-making. In this talk, we look at how these techniques can be
used in retail sales scenarios to enhance customer engagement, optimize strategies, and spur corporate growth.

ACKNOWLEDGEMENTS
We thank the employees and programmers of the Computer and Information Center at our beloved
university, Al-Balqa Applied University, for their cooperation and providing what is necessary to complete this
research. We also thank the administration of Ajloun University College for the support it provided throughout
the preparation of this scientific research. We can't forget our families for their patience and support.

REFERENCES
[1] M. Besta et al., “Demystifying graph databases: analysis and taxonomy of data organization, system designs, and graph queries,”
ACM Computing Surveys, vol. 56, no. 2, pp. 1–40, 2024, doi: 10.1145/3604932.
[2] Y. Shao and N. Nakashole, “On linearizing structured data in encoder-decoder language models: insights from text-to-SQL,” in
Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human
Language Technologies, 2024, pp. 131–156, doi: 10.18653/v1/2024.naacl-long.8.
[3] M. E. Coimbra, A. P. Francisco, and L. Veiga, “Study on resource efficiency of distributed graph processing,” arXiv-Computer
Science, pp. 1–23, 2017.
[4] A. Baudin, M. Danisch, S. Kirgizov, C. Magnien, and M. Ghanem, “Clique percolation method: memory efficient almost exact
communities,” in Advanced Data Mining and Applications, 2022, pp. 113–127.
[5] J. Kim, S. Lee, Y. Kim, S. Ahn, and S. Cho, “Graph learning-based blockchain phishing account detection with a heterogeneous
transaction graph,” Sensors, vol. 23, no. 1, 2023, doi: 10.3390/s23010463.
[6] X. Ren, K. Zhao, P. J. Riddle, K. Taskova, Q. Pan, and L. Li, “DAMR: Dynamic adjacency matrix representation learning for
multivariate time series imputation,” Proceedings of the ACM on Management of Data, vol. 1, no. 2, pp. 1–25, 2023, doi:
10.1145/3589333.
[7] Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, and P. S. Yu, “A comprehensive survey on graph neural networks,” IEEE
Transactions on Neural Networks and Learning Systems, vol. 32, no. 1, pp. 4–24, 2021, doi: 10.1109/TNNLS.2020.2978386.
[8] H. Chen et al., “G-tran,” Proceedings of the VLDB Endowment, vol. 15, no. 11, pp. 2545–2558, 2022, doi:
10.14778/3551793.3551813.
[9] D. Lin, J. Wu, Q. Yuan, and Z. Zheng, “Modeling and understanding ethereum transaction records via a complex network
approach,” IEEE Transactions on Circuits and Systems II: Express Briefs, vol. 67, no. 11, pp. 2737–2741, 2020, doi:
10.1109/TCSII.2020.2968376.
[10] A. Pismerov and M. Pikalov, “Applying embedding methods to process mining,” in ACM International Conference Proceeding
Series, 2022, pp. 1–5, doi: 10.1145/3579654.3579730.
[11] Z. Yang, Y. Bi, L. Wang, D. Cao, R. Li, and Q. Li, “Development and application of a field knowledge graph and search engine
for pavement engineering,” Scientific Reports, vol. 12, no. 1, 2022, doi: 10.1038/s41598-022-11604-y.
[12] M. Wu, X. Yi, H. Yu, Y. Liu, and Y. Wang, “Nebula graph: An open source distributed graph database,” arXiv-Computer
Science, pp. 1–18, 2022.
[13] A. Ferhati, “Applying a label propagation algorithm to detect communities in graph databases,” M.Sc. Thesis, Department of
Computer Science & Engineering, University of Bergamo, Bergamo, Italy, 2022.
[14] S. Biswas, M. Bhattacharyya, and S. Bandyopadhyay, “Topological analysis on multi-scenario graphs: Applications toward
discerning variability in SARS-CoV-2 and topic similarity in research,” Transactions of the Indian National Academy of
Engineering, vol. 7, no. 1, pp. 365–374, 2022, doi: 10.1007/s41403-021-00306-y.
[15] H. Seiti, A. Makui, A. Hafezalkotob, M. Khalaj, and I. A. Hameed, “R.Graph: A new risk-based causal reasoning and its
application to COVID-19 risk analysis,” Process Safety and Environmental Protection, vol. 159, pp. 585–604, 2022, doi:
10.1016/j.psep.2022.01.010.
[16] A. B. Ammar, “Query optimization techniques in graph databases,” International Journal of Database Management Systems,
vol. 8, no. 4, pp. 1–14, 2016, doi: 10.5121/ijdms.2016.8401.
[17] M. Mohajer, “A graph-based platform for customer behavior analysis using applications’ clickstream data,” arXiv-Computer
Science, pp. 1–23, 2020, doi: 10.48550/arXiv.2002.10269.
[18] P. Mehrotra, V. Anand, D. Margo, M. R. Hajidehi, and M. Seltzer, “SoK: The faults in our graph benchmarks,” arXiv-Computer
Science, pp. 1–26, 2024.
[19] P. Wills and F. G. Meyer, “Metrics for graph comparison: A practitioner’s guide,” PLOS ONE, vol. 15, no. 2, Feb. 2020, doi:
10.1371/journal.pone.0228728.
[20] C. Lezcano and M. Arias, “Characterizing transactional databases for frequent itemset mining,” CEUR Workshop Proceedings,
vol. 2436, 2019.
[21] J. Sandell, E. Asplund, W. Y. Ayele, and M. Duneld, “Performance comparison analysis of ArangoDB, MySQL, and Neo4j: An
experimental study of querying connected data,” in Proceedings of the Annual Hawaii International Conference on System
Sciences, 2024, pp. 7760–7769.
[22] A. S. Reddy, P. K. Reddy, A. Mondal, and U. D. Priyakumar, “Mining subgraph coverage patterns from graph transactions,”
International Journal of Data Science and Analytics, vol. 13, no. 2, pp. 105–121, 2022, doi: 10.1007/s41060-021-00292-y.
[23] M. Lei et al., “Mining top-k sequential patterns in transaction database graphs: A new challenging problem and a sampling-based
approach,” World Wide Web, vol. 23, no. 1, pp. 103–130, 2020, doi: 10.1007/s11280-019-00686-w.
[24] Z. Yao, “Visual customer segmentation and behavior analysis: A SOM-based approach,” M.Sc. Thesis, Department of
Information Technologies, Åbo Akademi University, Turku, Finland, 2013.
[25] W. A. Alzoubi, “Dynamic graph based method for mining text data,” WSEAS Transactions on Systems and Control, vol. 15,
pp. 453–458, 2020, doi: 10.37394/23203.2020.15.45.
[26] A. Bóta and M. Krész, “A high resolution clique-based overlapping community detection algorithm for small-world networks,”
Informatica, vol. 39, no. 2, pp. 177–187, 2015.
Graph-based methods for transaction databases: a comparative study (Wael Ahmad AlZoubi)
1672  ISSN: 2252-8938

[27] S. Tabassum, F. S. F. Pereira, S. Fernandes, and J. Gama, “Social network analysis: An overview,” Wiley Interdisciplinary
Reviews: Data Mining and Knowledge Discovery, vol. 8, no. 5, 2018, doi: 10.1002/widm.1256.
[28] Z. Huang, S. Zhang, C. Xi, T. Liu, and M. Zhou, “Scaling up graph neural networks via graph coarsening,” in Proceedings of the
ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2021, pp. 675–684, doi:
10.1145/3447548.3467256.
[29] X. Liu et al., “Survey on graph neural network acceleration: an algorithmic perspective,” in IJCAI International Joint Conference
on Artificial Intelligence, 2022, pp. 5521–5529, doi: 10.24963/ijcai.2022/772.
[30] V. Yoghourdjian, Y. Yang, T. Dwyer, L. Lawrence, M. Wybrow, and K. Marriott, “Scalability of network visualisation from a
cognitive load perspective,” IEEE Transactions on Visualization and Computer Graphics, vol. 27, no. 2, pp. 1677–1687, 2021,
doi: 10.1109/TVCG.2020.3030459.
[31] M. Hlawatsch, M. Burch, and D. Weiskopf, “Visual adjacency lists for dynamic graphs,” IEEE Transactions on Visualization and
Computer Graphics, vol. 20, no. 11, pp. 1590–1603, 2014, doi: 10.1109/TVCG.2014.2322594.
[32] S. Zhang, H. Tong, J. Xu, and R. Maciejewski, “Graph convolutional networks: a comprehensive review,” Computational Social
Networks, vol. 6, no. 1, 2019, doi: 10.1186/s40649-019-0069-y.
[33] I. Amaral, “Complex networks,” in Encyclopedia of Big Data, Cham: Springer International Publishing, 2022, pp. 198–201.
[34] H. Xuanyuan, P. Barbiero, D. Georgiev, L. C. Magister, and P. Liò, “Global concept-based interpretability for graph neural
networks via neuron analysis,” Proceedings of the 37th AAAI Conference on Artificial Intelligence, AAAI 2023, vol. 37, no. 9,
pp. 10675–10683, 2023, doi: 10.1609/aaai.v37i9.26267.
[35] H. Rawlani, “Visual interpretability for convolutional neural network,” Towards Data Science, pp. 1–20, 2018.
[36] M. Li, Y. Deng, and B. H. Wang, “Clique percolation in random graphs,” Physical Review E - Statistical, Nonlinear, and Soft
Matter Physics, vol. 92, no. 4, 2015, doi: 10.1103/PhysRevE.92.042116.
[37] I. R. Ward, J. Joyner, C. Lickfold, Y. Guo, and M. Bennamoun, “A practical tutorial on graph neural networks,” ACM Computing
Surveys, vol. 54, no. 10, pp. 1–35, 2022, doi: 10.1145/3503043.
[38] B. Khemani, S. Patil, K. Kotecha, and S. Tanwar, “A review of graph neural networks: concepts, architectures, techniques, challenges,
datasets, applications, and future directions,” Journal of Big Data, vol. 11, no. 1, 2024, doi: 10.1186/s40537-023-00876-4.
[39] S. Dutta and S. Roy, “Complex network visualisation using JavaScript: a review,” in Intelligent Systems, vol. 431, 2022, pp. 45–53.

BIOGRAPHIES OF AUTHORS

Wael Ahmad AlZoubi holds a doctor of computer sciences from National

University of Malaysia in 2013. He also received his B.Sc. and M.Sc. (computer science) from
Yarmouk University, Jordan in 2000 and 2004, respectively. He is currently an Assistant
Professor at Department of Computer Science in Al-Balqa Applied University, Ajloun,
Jordan. His research includes meta-heuristics, global optimization, machine learning, data
mining, bioinformatics, graph theory and parallel programming. He has published over 20
papers in international journals and conferences. He can be contacted at email:
[email protected].

Dr. Ibrahim Mahmoud Alturani is an instructor in the Department of Computer

Science at Ajloun College, Al-Balqa Applied University, Jordan. He earned his B.S. and M.S.
degrees in computer science from Yarmouk University, Jordan, in 2004 and 2007,
respectively, and completed his Ph.D. in computer science at the University Malaysia
Terengganu, Malaysia, in 2021. He began his academic career as a part-time lecturer in the
Department of Computer Science at Yarmouk University from 2007 to 2008 before joining
Al-Balqa Applied University as an instructor, where he has been teaching since 2008. He has
published several papers in international journals, with research interests encompassing
knowledge representation through ontology and knowledge graphs, natural language
processing, content-based retrieval, and artificial intelligence. He can be contacted at email:
[email protected].

Roba Mahmoud Ali Aloglah received her bachelor's degree of information

technology from Al-Balqa Applied University in 2004. She received the master's degree from
the Arab academy Jordan, Amman in 2005. She is a lecturer of computer science and
information technology at Department of Management Information Science, Amman College
for Financial and Managerial Sciences, Al-Balqa Applied University, Amman, Jordan since
2008. Her research interests include algorithms, computer networks, artificial intelligence and
computer security. She can be contacted at email: [email protected].

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1663-1672