support, Automated IT support systems, NLP .

Effective communication serves as the cornerstone of every successful process, whether human -driven or automated. The ability to comprehend one another and exchange ideas and information facilitates collaboration towards achieving common objectives. This h olds especially true in the realm of Information Technology (IT) Services , where clear communication is essential for articulating needs and issues and enabling colleagues to grasp the intricacies of requests thoroughly. A detailed and semantically accurat e description increases the likelihood of the message recipient interpreting it correctly and delivering an appropriate solution. Ultimately, the overarching aim of this scenario is to resolve issues or requests efficiently and effectively.

A ticket is the first step in any Service Desk occurrence. An incident, problem, or service request is an example of a service event, and a ticket is a historical record that describes it. The processing of a service event is governed and controlled by tickets. They are employed in the routing of events for resolution among various resources [1] .

The Iran Food and Drug Admini stration (IFDA) fulfils an essential part in ensuring public health by enforcing laws in a variety of industries, including food, cosmetics, and medicines. The obligation emphasizes the need for a strong infrastructure that can handle inquiries for assista nce from companies looking for regulatory clearance or direction. But despite the recent paperless system upgrade having improved productivity, there has been a noticeable increase in the number of electronic tickets issued. The IFDA's current workforce fa ces an important obstacle as a result of the rise in requests, which might result in longer Response Times (RT) and delays in crucial regulatory processes.

A ticket is the first step in the world of Service Desk incidents. It acts like an archive that defi nes service events by collecting events, difficulties, or service requests. Tickets control the flow of activities among multiple resources and are the basis for the systematic processing of service events [1] . Although they often have helpful information, tickets are typically disorganized. In theory, in order to understand the extra capabilities, new members should review the most important tickets.

On the other hand, this procedure takes a while. Even though tickets are frequently processed one at a time, groups of tickets occasionally show comparable trends [2] . The digitalization approaches implemented by all sectors have resulted in a notable increase in the volume of support requests [3] . A vital tool for successfully and efficiently handling client requests and concerns, Ticket Management Systems (TMS) are utilize d by a wide range of companies and associations.

Resolution Time (RT), which is the period that occurs between the time that an issue is logged and when it is fully resolved, is a measure of multiple challenges that modern Customer Relationship Management (CRM) systems encounter. Each of these issues can be overcome with Real -Time Prioritization (RTP), especially when it comes to improving the support ticketing module. Companies are now able to assign customer support interactions to personnel, rank them, a nd efficiently follow their growth [4] .

7 6/3

3/2 3/1

Pharmaceuticals Cosmetics & Supplements & Food Control Medical Food and Hygiene Natural Products Laboratories Equipment Beverages

The average response time (in days) for questions sent to the several IFD A departments is shown in Fig. 3 .

The departments are classified on the x -axis, and the average response time is measured on the y -axis. Among all the specialities shown in the chart, the Department of Food and Beverage and the Department of Cosmetics and Hygiene stand out for having the longest resolution timeframes. The cause for this prolonged response time is probably due to a great deal of questions that these business large number of companies generate. However, the IFDA cannot simply incre ase staff to handle the increasing volume of inquiries due to financial constraints. This requires a more systematic and productive method. The IFDA is looking at using Machine Learning (ML) and Artificial Intelligence (AI) to develop an original approach to ticket -setting priorities, having recognized the potential benefits of these technologies.

There has never been a greater need to use modern technology to automate regulatory processes, especially given the mistakes that are common in data collecting an d use. This change offers a solution to the growing issue of regulatory organizations' dwindling internal staffing levels along with simplifying operations [7] .

This study contributes to the ongoing pursuit of enhanced customer support experiences. By leveraging Tag Cloud Representation for ticket prioritization, IFDA can expedite issue resolution, improve resource allocation within its IT service desk, and ultimately enhance customer satisfaction. Furthermore, the research emphasizes the importance of continuous improvement . The exploration of additional ML algorithms alongside NLP presents a promising avenue for further optimizing the digitalization of IFDA's customer support system, leading to a more robust and efficient service experience for its customers. Additionally, the main re search question of this study: how can customer support systems be optimized in a way that responses to submitted tickets are faster and more effective? is practically addressed in Section 4.

The rest of the paper is structured as follows. Secti on 2 provides the literature review, and in Section 3, the research methodology is explained. Section 4 provides the answer to the main research question (data analysis ). Finally, the conclusion and policy implications of the paper are stated in Section 5.

Any business, given the current level of digitization, has a wide range of applications, many of which have grown continuously. Massive and complex IT service environments are needed to serve such a portfolio [8] .

These advancements highlight the critical role that IT support systems play in the support functions of any firm [9] . Tickets serve as the most significant means for interaction between clients and the personnel in charge of monitoring service, enabling the resolution of any issues or events. These kinds of deali ngs occur in almost every sector of the economy. However, complaints and requests for assistance for IT -related issues are the most common examples [1] . Ticket Automation (TA) can be defined as the collection of automated systems that aim to reduce the number of steps between the submission of a ticket and its resolution. The rising complexity of IT services us ed by businesses has drawn attention to the automatic classification of IT issues [10] .

The initiation of a problem ticket within the helpdesk support system is carried out by the end user, who chooses a category and provi des a description. Nevertheless, the manual selection of the ticket category by the end user introduces the risk of misrouting tickets to incorrect resolver groups. The E -ticketing system, endorsed by the Federal Highway Administration's Every Day Counts (EDC) program, uses software tools to track electronically and archive data regarding the quantity of construction materials procured by State Transportation Agenci es (STAs) through unit bid contracts based on weight [11] .

The rapid evolution of technologies, particularly driven by contemporary AI, has significantly transformed the world, ideally to the advantage of humanity. AI is being applied to address numerous real -world challenges, benefiting society in various ways. Notable ex amples include industrial automation, intelligent virtual assistants, autonomous vehicles, cancer diagnostics, and advanced Enterprise Resource Planning (ERP) systems [12] .

The advent of ML has opened avenues for automating ticket classificati on, thereby facilitating the prediction of the time required to resolve cases [13] , [14] . Various ML algorithms have been integrated into Service Ticketing Systems (STS) to achieve this goal. In previous research, an in -depth linguistic analysis was conducted on ITIL Change Management (CHM) ticket texts from the IT CHM ticket processing department of a large enterprise with over 200,000 employees worldwide [10] . This study utilized detailed linguistic representations of the ticket texts to implement a rule -based approach for predicting process complexity [15] .

Several automatic document classification methods have been developed using supervised learning techniques, as documented in prior studies [16] . For instance, Revina et al. [17] proposed a method employing Support Vector Machines ( SVM) to detect commits carrying safety risks automatically. Diao and Bhattacharya [18] utilized Word2Vec and a Convolutional Neural Network (CNN) to classify functional and non -functional requirement documents. Additionally, Pingclasai et al. [19] introduced a technique fo r categorizing bug reports using supervised learning and topic modelling methods like Latent Dirichlet Allocation (LDA). However, it is worth noting that labelled training data preparation can be a costly endeavour. Moreover, in cases where the context of the software development project is not predetermined, the derivation and creation of labels pose significant challenges.

Al-Hawari andBarham [20]in a ML -based help desk system for ITSM was introduced to enhance the performance of the German Jordanian University IT staff in addressing technical ticket issues. This system employs a ML model to automatically classify tickets by utilizing the title, description, and comments of each ticket. Specifically, TF -IDF is applied for tokenization and feature extraction, while the SVM algorithm is used for the classification task.

A recent study introduced a SVM classifier that integrates various features such as Bag-Of-Words (BOW), POS -tag, synonyms, and entity types [21] . This research aligns closely with the objectives of our investigation, particularly regarding the extraction technique employed to analyze diverse features within each question.

However, it's noteworthy that the dataset utilized in the study is the Hierarchical Classification Standard, which differs from the focus of our research, centred on the da taset specific to helpdesk support systems.

Additionally, Rodríguez -Robayo et al. [22]introduced a Tree-Based Convolutional Neural Network (TBCNN) tailored for p rogramming language processing. A pivotal study, the National Cooperative Highway Research Program (NCHRP) Synthesis 450, sheds light on the significant discrepancy between the expanding scope of regulatory responsibilities and the diminishing workforce av ailable to fulfil them [23] , [24] .

The categorization of questions stands as a pivotal component within the realm of question processing as it determines the appropriate type of response. The classification of answers assumes significant importance in Question Answering Systems, delineating the specific information that needs to be retrieved from a knowledge base [25] .Assigning a text to a set of pre -defined categories is known as text classification, sometimes known as categorized text or language tagging. Human specialists have traditionally completed this type of task. For instance, expert text classif ication is still frequently used in qualitative studies such as indexing or coding [26] .However, as the volume of text data increased, scientists' and practitioners' emphasis turned to methods that were automated. Three primary categories of automatic text classification techniques exist rule -based, ML -based and hybrid systems that combine both of them. Software development and maintenance ticket classification have been investigated to address issues like precise ticket allocation and prioritizing, forecasting time and quantity of possible tickets, and pr eventing duplicate requests [27] , [28] .

Showcase the feasibility o f extracting tags possessing collective utility from the amalgamation of individually and freely assigned tags, thereby autonomously resolving tag ambiguity issues [29] . Various machine -learning techniques are explored for email filtering and classification. Additionally, SVM and Artificial Neural Networks (ANN) have been employed by other researchers [30] . Research indicates that the Neural Network (NN) footprint can be significantly reduced with only moderate performance degradation across various tasks such as image recognition [31] , natural language understanding and speech recognition. Various techniques have been proposed to achieve this, including integer quantization [32] , hashing, vector quantization [33] , and matrix factorization. N N models often need to be executed rapidly to minimize user -perceived latency. This is particularly crucial for applications involving continuous interactions with customers, such as virtual assistants [34] .

Delivering top -notch customer service is paramount in both service -based industries and businesses. It's imperative t o prioritize customer satisfaction by providing optimal assistance. The IT service delivery of every company heavily depends on the efficiency of its helpdesk. Consequently, numerous enterprises employ intelligent helpdesk solutions to enhance the quality of customer service, recognizing its critical importance [35] . A customer's request for assistance from a service p rovider's help desk is r eferred to as a support ticket .

These are necessary resources for all contemporary companies to manage their interaction with clients and include incident reports, service tickets, and customer complaints [11] .

In recent years, the adoption of helpdesk support systems has emerged as a prevailing trend across industries, offering a structured approach to managing support services. A plethora of open -source applications catering to helpdesk support has already permeated the market, providing organizations with readily available solutions to facilitate their support endeavours. Management can peruse these options and select the most suitable helpdesk support system based on the unique requirements and objectives of their business operations [13] .

The service desk creates a ticket once it recognizes the situation as an incident. The incident report, the user's name and contact information, the date and time of the occurrence, and other details should all be included in the ticket. Categorization, ordering, and the actions the service desk takes may all be included in the logging process. The process of categorization entails giving the occur rence of at least one subcategory and a category [21] .

In a helpdesk system, a crucial element is a Question and Answering (Q&A) system, housing question data that serves as the primary information for processing to ensure ef fective implementation of Q&A activities.

Hence, there is a necessity to organize existing questions to facilitate accurate and prompt responses. A question, in this context, refers to a linguistic expression utilized to seek information [36] .

Numerous methods for categorizing questions have been explored and put into practice, particularly within customer service domains. Three primary strategies are commonly employed for question classification tasks:

these include 1) ML approaches, 2) rule -based techniques, and 3) hybrid methodologies that combine elements of both [39].

Pingclasai et al. [19]conducted a comparative analysis of four algorithms: 1) Multinomial Naïve Bayes, 2) Logistic Regression, 3) K -nearest Neighbor, and 4) SVM . Their findings indicated that the SVM exhibited the highest accuracy rate of 87% when compared to Logistic Regression (81%), Multinomial Naïve Bayes (69%), and K-nearest Neighbor (67%) on the training dataset. Notably, the SVM classifier demonstrated consistent performance across various service desk ticket data samples, achieving a satisfactory level of accuracy. The implementation of the suggested automated ticket classifier system promises enhancements in end -user satisfaction, customer happiness, and the efficient utilization of support resources. Additionally, it facilitates faster tick et RT s and f osters business growth.

[11] Clustering Applied clustering techniques to tag cloud design .

Alphabetical, semantic, and random [37] Alphabetical design was the best rated .

designs Users can manipulate typography, position, orientation, [22] Manipulable design and color of tags .

[28] Tag locatio n (first quadrant) Tags in the first quadrant are most remembered .

Table 1 summarizes various design techniques used in tag cloud research and the key findings associated with each approach. It highlights the impact of different designs on user efficiency, memory, and interaction with tag clouds.

As highli ghted at the outset of this study, the growing integration of AI into business and daily life underscores the critical importance of AI explainability. Understanding AI system decisions is crucial for building trust and effectively addressing errors made b y such systems. Our research focuses on the ITIL Change Management-based IT ticketing process of a large telecommunications company. Implementing changes in the IT infrastructure involves interventions in the organization's IT environment, functions, and u ser experience each time a change is made.

This section of the article delves into the analysis of the employed methods and addresses the posed inquiries to present a practical and effective approach for addressing the research problem. Leveraging ML algorithms, this section explai nsthe methods that were used and gives practical solutions to the challenge that was explored. By examining actual cases in the appropriate field, these evaluations give investigators the chance to identify trends and subsequently acquire expertise for cl osing the understanding gap between theory and practice. Therefore, the practical strategy used in this part helps to improve understanding and effectiveness while solving real -world problems in the sector.

As we became familiar with the systems related to the Food and Drug Administration in the previous section, we will now, considering the various subsystems under the TTAC system for transferring implementation experiences in the regulatory industry in the health sector, randomly implement and introduce o ne of the targeted systems called Entities .

In the Entitysystem, by executing the Request for Technical Manager Qualification operation by the applicant, a task list is created for them. The technical manager information registration task list is specifically designed for completing the personal information, educational, and wo rk history of technical managers.

Additionally, the registration task list for the CEO is specifically designed for registering legal companies active in the health sector. This task list includes the personal information of the CEO and board members, along with the registered company information . After reg istering and submitting the technical manager's information in the system, the information review process is carried out by the Food and Drug Administration experts, and if the information is approved, an activity license for the technical manager is issue d. After the activity license is issued, company managers can search for the relevant technical manager and send an employment request to them. The technical manager, by viewing the applicant companies' information in the system, can accept o r reject the e mployment requests.

In Fig. 6, to facilitate and expedite the response to registered tickets, we will categorize and classify them, and in the continuation of the article, we will discuss how this ca tegorization is used in the applied algorithms.

In this section, we provide a detailed solution to optimize customer support systems, addressing the main research question of this study:

Q1: How can customer support systems be optimized in a way that resp onses to submitted ticke ts are faster and more effective?

Given the heavy workload in the Ian Food and Drug Administration (IFDA), the presence of specialized departments with numerous requests, the submission of duplicate tickets, the outbreak of infectious diseases such as COVID-19, and several other factors, the need to improve the customer service support system became critically important. Support systems for users require precise prioritization to respond effectively and efficiently to submitted tickets. To address this, we conducted group meetings with experts from specialized departments to gather and document their feedback. Subsequently, in collaboration with IT engineers and utilizing the Tag Cloud R epresentation technique, the status of open tickets in the system was graphically displayed to the responsible experts in a structured format.

The Tag Cloud Representation technique helps categorize tickets based on frequently used keywords and similar topics. This meth od enables experts to identify high -priority tickets and respond appropriately quickly.

Additionally, by displaying frequently asked questions in the users' dashboard, duplicate ticket submissions are prevented, thereby reducing the workload.

Using this te chnique resulted in:

I. Increased efficiency and productivity: experts were able to respond faster to more important tickets with proper prioritization.

II. Reduction in duplicate tickets: by displaying frequently asked questions to users, the number of duplicate tickets decreased, reducing the workload for experts.

III. Improved user satisfaction: rapid and effective responses to tickets led to higher user satisfaction.

IV. Better resource management: with fewer duplicate tickets, the organization's resources wer e managedmore optimally.

Statistics indicate that the implementation of this method has led to a decrease in the number of submitted tickets and an improvement in the quality of responses. These measures have significantly enhanced the efficiency and red uced the workload of the support units. All the code related to this section, written using the Jupyter framework, has been included in the attached file submitted with this paper for the journal.

Average response time (before)Average response time (after)

7 6/3 6 5/2 5 4/3 4 3/2 3/3 3 2/2

Pharmaceuticals Cosmetics & Supplements & Food Control Medical Food and Hygiene Natural Products Laboratories Equipment Beverages

The limitations identified across studies addressing behaviour pattern analysis in ticketing systems span several aspects. These include challenges such as noisy data affecting pattern recognition, limitations in handling technical terms impacting accuracy , algorithmic constraints hindering ticket classification, computational resource requirements for expedited learning, the need for periodic training set adjustments, absence of ground truth annotations for sentiment analysis, dataset constraints, the nece ssity to evaluate system performance over longer time scales and under varying settings, and the initial handling of unstructured noisy data. Addressing these limitations collectively would require strategies ranging from data preprocessing techniques to a lgorithmic enhancements and resource optimization to ensure accurate and efficient behaviour pattern analysis in ticketing systems.

This paper explores the limitations and potential solutions for behaviour pattern analysis in ticketing systems. By analyzin g user interactions within ticketing systems, organizations can gain valuable insights into customer behaviour, identify potential issues, and improve overall service delivery. However, several challenges hinder the accuracy and efficiency of current behav iour pattern analysis methods.

This investigation points up a number of major shortcomings in previous research on behavior pattern analysis in ticketing systems. Some of these challenges are:

I. Noisy data: ticketing system data can be riddled with errors an d inconsistencies, impacting pattern recognition algorithms [14] .

II. Technical terminology: the presence of technical terms within tickets can hinder the accuracy of analysis, as classification algorithms may not readily understand these ter ms.

III. Algorithmic constraints: current classification algorithms might struggle to effectively categorize tickets due to inherent limitations in their design [1] .

IV. Computational resources: training and running complex algorithms often require significant computational resources, impacting processing speed and scalability.

V. Training set maintenance: the performance of behaviour pattern analysis systems can degrade over time if the training data sets are not periodically adjusted to reflect evolving user behaviour and system configurations.

VI. Sentiment analysis challenges: the absence of ground truth annotations for sen timent analysis within ticketing data makes it difficult to assess us er emotions and satisfaction accurately [40] .

VII. Dataset constraints: dataset limitations, such as size and representativeness, can restrict the generalizability of findings and hinder the development of robust behaviour pattern analysis models.

By addressing these challenges and implementi ng the proposed solutions, researchers and practitioners can develop more accurate and efficient behaviour pattern analysis systems for ticketing systems. This will ultimately lead to improved customer service delivery and enhanced decision -making within o rganizations.

The provision of support services throughout the entire application system development process stands as a cornerstone for organizational success. Both technical and management support play indispensable roles in guaranteeing the seamless evo lution and ultimate triumph of an application system. From the initial user requirement phase through programming development, system testing, and culminating in the final support stage, ongoing assistance is imperative to shepherd the system to its full m aturity. Indeed, the continuity of support services is essential for sustaining the relevance and functionality of an application system over time . A ML system for categorization and ranking incoming calls can be developed by exploiting the IFDA's huge col lection of previous inquiries and replies. The platform may evaluate incoming inquiries and assign priority levels depending on public health impact and urgency by developing models on earlier data. This ensures prompt attention to critical issues, optimiz ing resource allocation and regulatory decision -making. When an ML -based ticket prioritizing system is effectively implemented at the IFDA, it can optimize the utilization of resources, simplify workflows, and greatly speed up RT for severe public health c oncerns. By using ML to find patterns in questions, human specialists can focus on deeper issues, which increases productivity and eliminates backlogs.

The IFDA has strong databases that are used to publicly label ML algorithms like K -Nearest Neighbours (KNN) and others. This enables improved performance comparison and exploitation. Moreover, through the development of a multi -year strategy and the setting up of appropriate performance measures, the Support Ticketing system can be improved by giving priorit y to ongoing improvement.

The high volume of inquiries encountered in the transitioned paperless environment at the IFDA necessitates a novel solution. Implementing a ML system for ticket prioritization offers a promising approach. By leveraging its vast d ataset of past inquiries and resolutions, the IFDA can develop a system capable of intelligently classifying and prioritizing incoming tickets based on urgency and potential public health impact. This innovative solution holds the potential to enhance publ ic health protection, optimize resource allocation within the IFDA, and improve customer service for businesses seeking regulatory guidance. Further R&D are crucial to refine this approach and ensure its successful integration within the IFDA's regulatory framework.

The conceptualization and ideas for the study were developed by H.S. The data collection and formal analysis were also carried out by H.S. Funding for the research was acquired by A.H. Both A.H. and H.S. were involved in the investigation phase. Methodology development, software programming, and the creation of computer code and supporting algorithms were handled by H.S. Validation and review were conducted by A.H., while the original draft of the manuscript was written by H.S. The review and editing process was completed by A.H.

The data used in this research study is derived from th e ticketing system of the IFDA . The FDA retains the ownership of this data. Given the nature of the data, which includes inf ormation related to the production lines of certain pharmaceuticals and specific drug -related details, sharing this data poses significant risks and potential harm. Despite our trust in your esteemed journal, we must adhere to these confidentiality constra ints. However, we have included the relevant Python software codes in the appendix for the journal's review.

The author declares no conflict of interest. The data utilized in this study were obtained from the ticketing system of the IFDA . Access to this data was granted under strict confidentiality agreements, and the IFDA retains full ownership of the data. Therefore, the data cannot be shared publicly or with third parties without explicit permission from the IFDA.