Customer Support Ticket Categorization and Prioritization Using

Natural Language Processing

C. S Kanimozhi Selvi, S. Jyothi Shri, R. Prasshanthini, Neelamegan and R. Sanjay

Department of Artificial Intelligence, Kongu Engineering College, Erode, Tamilnadu, India

Keywords: Customer Support, NLP Automation, Ticket Categorization, Prioritization Models, Supervised Learning,

Deep Learning.

Abstract: Efficient and effective customer service relies heavily on the systematic management of customer support

tickets. This paper presents an Natural Language Processing (NLP) - driven system designed to automate the

categorization and prioritization of customer support tickets, significantly enhancing service efficiency. The

system analyzes ticket content using machine learning models and deep learning models for both

categorization and prioritization tasks. Categorization is achieved using Topic Modeling (NMF) to identify

department-specific categories, while ticket priority levels are determined by extracting urgency and impact

keywords. By leveraging these techniques, the system streamlines ticket handling, reduces manual

intervention, and optimizes resource allocation. Experimental results demonstrate high accuracy, scalability,

and improved operational efficiency, ultimately enhancing customer satisfaction.

1 INTRODUCTION

In today’s highly competitive market, customer

support is the order of the day for customer

satisfaction and loyalty. Support teams face the

ongoing challenge of managing a high volume of

diverse queries, such as payment issues, general

inquiries, technical problems, and product requests.

Accurate categorization and prioritization are crucial

to ensure swift, personalized resolutions that meet

customer expectations. Traditional ticket handling

processes rely heavily on manual efforts, which are

time-consuming, labor-intensive, and prone to human

error. As customer bases expand, the need for

intelligent, automated solutions to streamline these

operations becomes increasingly evident. Efficient

categorization and prioritization are, therefore,

critical components of modern customer support

systems.

This study introduces an advanced framework

leveraging machine learning and deep learning

techniques to address these challenges effectively.

The framework classifies support tickets into

appropriate departmental categories and assigns

priority levels based on factors like urgency and

impact. Using a comprehensive dataset of 78,313

customer complaints, the work rigorously builds and

evaluates various ML and DL models for these tasks.

To handle unlabeled data, Topic Modeling techniques

such as Non-negative Matrix Factorization (NMF)

are applied to categorize queries into five distinct

departmental groups. By automating support ticket

management, this framework aims to significantly

enhance operational efficiency, reduce manual effort,

and improve overall customer satisfaction. This

automation ultimately strengthens customer

relationships and fosters long-term loyalty by

providing timely, accurate support.

2 LITERATURE SURVEY

In 2024, Carla Vairetti et al. (Vairetti, Aránguiz, et al.

, 2024) introduced a framework that combines

multicriteria decision-making (MCDM) and deep

learning to prioritize complaints. The framework

automates complaint classification and prioritization

using text analytics and operational research

techniques. It bridges deep learning with traditional

machine learning by incorporating pretrained models.

MCDM is applied to combine multiple criteria into a

single prioritization score, enhancing decision-

690

Selvi, C. S. K., Shri, S. J., Prasshanthini, R., Neelamegan, and Sanjay, R.

Customer Support Ticket Categorization and Prioritization Using Natural Language Processing.

DOI: 10.5220/0013640400004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 3, pages 690-698

ISBN: 978-989-758-763-4

making. The framework's effectiveness is

demonstrated by its application to real-world data

from a Chilean agency. This approach improves

efficiency in managing complaints across various

service sectors.

In 2022, Alvaro Aldunate et al. (Aldunate,

Maldonado, Declerck, 2022)presented a

methodology which utilizes BERT and transfer

learning to uncover customer satisfaction factors

across various sectors. It highlights the superiority of

deep learning models over traditional text mining

methods in classification accuracy. The research

emphasizes the importance of automated evaluation

of consumer feedback for decision-making in the

service sector. A four-step methodology is used to

extract relevant insights from open-ended survey

responses. The study demonstrates how NLP and

deep learning can enhance customer experience

analysis.

In 2023, Peter Adebowale Olujimi et al. (Olujimi,

and, Ibijola, 2023)] reviewed 73 studies on the use of

NLP to automate customer queries across various

industries. The research highlights benefits like faster

response times, improved accuracy, and higher

customer satisfaction, indicating the increasing

demand for automated customer care systems. It

suggests that future research could explore advanced

NLP models and AI integration to further enhance

consumer interactions. The study emphasizes the

importance of thorough literature reviews for

credibility and adds to the growing body of

knowledge on NLP’s role in transforming customer

service and boosting corporate performance.

In 2021, H. A. Ahmed et al. (Ahmed, Bawany, et

al. , 2021) introduced CaPBug, a system for

automating software bug prioritization and learning

models, including CNNs, BiLSTMs, and BERT. It

demonstrated that BERT combined with TF-IDF and

Logistic Regression achieved the best macro-

averaged F1-score, highlighting the effectiveness of

pre-trained models in automating complaint

prioritization and improving resource allocation in

customer service.

The system analyzes bug reports from Eclipse

and Mozilla using supervised machine learning and

natural language processing (NLP). Bug reports were

manually classified into six categories and five

priority levels from 2016 to 2019. The system

predicts bug categories and priorities using textual

and categorical data, with feature extraction using

TF-IDF and NLP algorithms. Popular classification

methods are employed. The system improves

software maintenance by accurately predicting issues

and addressing class imbalance in priority levels.

In 2020, Nikhil Patel et al. (Patel, and, Trivedi,

2020) explored the use of predictive modeling,

machine learning, NLP, and AI chatbots to enhance

customer support and loyalty. The study examines

NLP applications across industries like marketing, e-

commerce, healthcare, telecommunications, and

finance, based on 26 articles from 2015 to 2022.

Common techniques like TF-IDF and SVM are

highlighted. The research discusses the need for

larger datasets to improve NLP applications in

customer service. Various models like TextCNN,

AdaBoost, and LDA were also applied in the study,

focusing on response quality, helpfulness, and

appropriateness.

In 2018, Sridhar Ramaswamy et al. (Ramaswamy,

and, DeClerck, 2020)explored using NLP and deep

learning for customer perception analysis. The study

integrates various technologies to extract insights

from consumer feedback, emphasizing the

importance of understanding consumer attitudes and

preferences. It discusses methods like named entity

recognition, rule-based semantics, and semantic

annotation to improve perception analysis. The

authors suggest using industry-specific survey

questions for more detailed insights.

In 2018, Ruanda Qamili et al. (Qamili, Shabani, et

al. , 2018) aimed to enhance customer service

productivity by incorporating machine learning into

ticketing systems. The study covers sentiment

analysis, ticket assignment, and spam detection in

customer support. It proposes an automated solution

to improve ticket management and reduce false

positives in spam filtering using a conservative

unanimity method. The authors emphasize the

importance of addressing delayed issue resolutions in

customer service.

In 2021, Nokudaiyaval G et al. (Kirthiga, and,

Ghayathri, 2021)developed a system using NLP and

the BERT algorithm to reduce labor and save

customers' time in customer support. NLP is used for

speech recognition, while BERT handles text

classification and prediction. Unlike existing systems

using IVR to route calls, the proposed solution

generates automated responses without human

intervention. The system is pre-trained using a closed

dataset, with tokenization applied to customer input.

BERT’s bidirectional search locates key interaction

information, providing a response between start and

finish parameters. This approach improves customer

support services by reducing reliance on human

interactions.

In 2022, Blümel and Zaki (Blümel, and, Zaki,

2022) conducted a comparative analysis of classical

and deep learning-based natural language processing

Customer Support Ticket Categorization and Prioritization Using Natural Language Processing

691

methods to prioritize customer complaints. The study

combined feature engineering techniques like TF-

IDF and Word2Vec with machine learning classifiers

and deep learning models, including CNNs,

BiLSTMs, and BERT. It demonstrated that BERT

combined with TF-IDF and Logistic Regression

achieved the best macro-averaged F1-score,

highlighting the effectiveness of pre-trained models

in automating complaint prioritization and improving

resource allocation in customer service.

In 2018, Silva et al. (Silva, Pereira, et al. , 2018)

introduced a machine learning-based module to

automate the categorization of IT incident tickets.

The proposed system uses a support vector machine

(SVM) to classify tickets, achieving an accuracy of

approximately 89% on real-world data. The module

enhances incident management productivity by

reducing routing errors and minimizing delays in

ticket assignment, contributing to improved IT

service delivery and customer satisfaction.

In 2023, Reddy and Prabhu (Reddy, and, Prabhu,

2023) conducted a comprehensive review on the role

of artificial intelligence in request management

processes. The study highlights AI's transformative

potential in automating request capture, logging,

categorization, and prioritization using techniques

like NLP, machine learning, and chatbots. It discusses

the challenges in manual request management and the

benefits of AI in enhancing workflow optimization,

resource allocation, and scheduling. The paper

underscores the impact of AI in improving efficiency,

accuracy, and customer satisfaction in request

management systems.

In 2019, Lucini et al. (Lucini, Tonetto, et al. ,

2019) proposed a text mining framework to analyze

airline customer satisfaction using Online Customer

Reviews (OCRs). They examined 55,775 reviews

across 419 airlines, identifying 27 satisfaction

dimensions and 882 adjectives through Latent

Dirichlet Allocation (LDA). The framework achieved

a prediction accuracy of 79.95% for customer airline

recommendations. Key satisfaction factors included

cabin staff, onboard service, and value for money.

The study provides actionable insights, highlighting

the need for tailored strategies in customer service

and comfort based on cabin class preferences to

enhance competitiveness in the airline industry.

In 2022, Ishizuka et al. (Ishizuka, Washizaki, et

al. , 2022) proposed a novel method to improve

feature comprehension in software development

projects using issue tickets. The method categorizes

tickets through clustering and visualizes them using

heatmapping and principal component analysis

(PCA). It also includes ticket lifetime visualization

for time-series analysis and keyword relationships

among ticket categories. A case study on an industrial

project demonstrated its effectiveness in helping

project members and newcomers understand

implemented features. The study emphasizes the

value of structured, visualized tickets in enhancing

onboarding and comprehension of multi-dimensional

requirements in evolving projects.

In 2019, Al-Hawari and Barham (Hawari, and,

Barham, 2019) introduced a machine learning-based

help desk system aimed at improving IT service

management. The system utilizes a ticket

classification model to streamline ticket resolution by

automatically associating help desk tickets with the

correct service. The methodology involves training

ticket data, preprocessing, stemming, feature

vectorization, and algorithm tuning. Experimental

results revealed that incorporating ticket comments

and descriptions significantly improved model

accuracy from 53.8% to 81.4%. The system also

features administrator and user views, supports

automatic email notifications, and enables

performance measurement through key performance

indicators (KPIs) for IT staff and processes.

In 2023, Benitez Pereira et al. (Pereira, Pizzio, et

al. , 2023) presented a machine learning model for

classifying IT support tickets to enhance help desk

operations. The model categorizes incoming support

tickets into seven topics with an average precision of

75%. To support daily operations, a web prototype

was developed, offering both frontend and backend

functionalities for IT analysts. The code, model, and

anonymized data were made publicly available for

replication of the study. This approach helps

minimize ticket resolution time and improves user

satisfaction by automating ticket topic identification.

3 PROPOSED METHODOLOGY

3.1 Data Collection

The dataset comprises complaint texts submitted by

customers to a financial institution. This dataset,

sourced from Kaggle, is specifically designed to

facilitate the classification and prioritization of

customer support tickets. It contains a total of 78,313

individual complaints, each in JSON format,

capturing various issues and grievances reported by

users. The data aims to support the development of

models or systems that can categorize these

complaints effectively and determine their priority

levels, enhancing customer service workflows.

INCOFT 2025 - International Conference on Futuristic Technology

692

Figure 1: Department classification ML model’s accuracies.

3.2 Data Preprocessing

The initial phase of classifying customer complaints

into departments involves comprehensive text

preprocessing to ensure data quality and relevance.

This process begins by removing unnecessary or

duplicate fields from the dataset, retaining only

essential information for analysis. Blank or

incomplete entries are identified and eliminated to

maintain data integrity. Text data is standardized

through methods like converting all text to lowercase

and removing punctuation using regular expressions,

which simplifies further analysis. Lemmatization is

applied to reduce words to their root forms, enhancing

consistency and aiding tasks such as information

retrieval and sentiment analysis. Masked personal

data is removed to eliminate personal details of the

user.

Feature extraction is initially performed using

TF-IDF (Term Frequency-Inverse Document

Frequency), a technique that highlights the most

important and relevant terms within the complaints.

By analyzing the frequency of terms across multiple

documents, it helps identify key words that are

significant in the context of the dataset. These

extracted features are then used for topic modeling,

where the most common terms present across the

complaints are analyzed to uncover underlying, latent

themes within the text data. Topic modeling is carried

out using NMF (Non-negative Matrix Factorization).

For each topic, the top 10 words are extracted, and

these words are manually mapped to predefined

categories such as Retail Banking Operations, Credit

Card Management, Payment and Billing, Dispute

Reporting, and Mortgages/Loans. This mapping

process ensures that each customer complaint is

accurately classified into the appropriate department,

which facilitates better message forwarding and

streamlined management of the complaints. For

prioritization of complaints, urgency is determined by

matching text tokens against a predefined list of

urgent words that signify a time-sensitive issue.

Advanced NLP tools, such as spaCy and negspacy,

are used to detect negation within the text, ensuring

that phrases like "not urgent" are correctly

interpreted. This method guarantees that complaints

are accurately classified as urgent or non-urgent, even

when negation modifies the meaning of key words.

This approach enhances the ability to prioritize

critical issues effectively, ensuring that the most time-

sensitive complaints are addressed promptly.

Customer Support Ticket Categorization and Prioritization Using Natural Language Processing

693

3.3 Machine Learning Models

3.3.1 Logistic Regression

This algorithm establishes a relationship between the

input features and the probability of a binary outcome

by applying a logistic function, making it highly

effective for predicting outcomes in two categories.

This algorithm is ideal for classifying customer

complaints into predefined categories and prioritizing

them. In this case, it helps to extract the most relevant

terms from the complaints using TF-IDF, which

boosts the model’s ability to accurately categorize

and prioritize complaints. The model achieved a

classification accuracy of 91.99%, efficiently sorting

complaints into their respective categories. For

prioritization, the model achieved an accuracy of

90.11%, demonstrating its ability to accurately

determine the urgency of complaints.

3.3.2 Decision Tree

The Decision Tree is a algorithm which divides the

dataset into smaller subsets by selecting decision

nodes based on the most relevant features, forming a

tree structure where each leaf represents a

classification label or a continuous value. This

method is effective for handling complex datasets

with multiple features, and it is highly interpretable.

By creating decision boundaries based on feature

values, it excels at categorizing complaints and

prioritizing them according to urgency. The model

achieved a classification accuracy of 78.44%,

accurately sorting complaints into predefined

categories. For prioritization, it achieved an

exceptional accuracy of 99.96%, effectively

differentiating between urgent and non-urgent

complaints with high precision.

3.3.3 Random Forest Classifier

This generates a collection of decision trees, each

built from various subsets of the data, and then

averages their outputs to mitigate overfitting and

improve accuracy. This method is particularly

effective for handling complex datasets, managing

large volumes of data, and providing feature

importance metrics. By using an ensemble approach,

Random Forest reduces both bias and variance

compared to a single decision tree, making it a

reliable choice for classifying and prioritizing

complaints. The model achieved a classification

accuracy of 81.13%, successfully organizing

complaints into predefined categories. For

prioritization, it reached an accuracy of 90.03%,

effectively distinguishing between urgent and non-

urgent complaints.

3.3.4 Support Vector Machine

Support Vector Machine (SVM) identifies the

optimal hyperplane that separates data points into

distinct classes, maximizing the margin between the

nearest points of different categories. This makes

SVM particularly effective for handling high-

dimensional datasets and complex classification

problems. Additionally, it excels in scenarios where

the data is not linearly separable, a common

characteristic of real-world datasets. In this study,

SVM achieved an impressive classification accuracy

of 91.51%, effectively sorting customer complaints

into predefined categories. For prioritization, the

model demonstrated a strong performance with an

accuracy of 94.68%, efficiently distinguishing

between urgent and non-urgent complaints.

3.3.5 Multinomial Naïve Bayes

Multinomial Naive Bayes is a algorithm tailored for

text classification tasks. It leverages Bayes' Theorem

and operates under the assumption that features, such

as words in a text dataset, are conditionally

independent of each other when conditioned on the

class label. This algorithm is ideal for classifying

customer complaints into predefined categories and

prioritizing them, as it can handle large vocabularies

and text-based features efficiently. Its probabilistic

nature helps in assigning the most likely category

based on the frequency of words in the complaints.

The model achieved an accuracy of 71.87% for

classification, accurately categorizing customer

complaints into predefined categories. For

prioritization, the model achieved an accuracy of

89.39%, although it performed poorly on the "urgent"

category, as reflected in its recall of 0%.

3.3.6 Gradient Boosting Machines

Gradient Boosting Machines (GBM) is an ensemble

algorithm that integrates multiple weak models to

form a robust predictive system, making it well-suited

for complex tasks such as text classification. It has

proven effective in categorizing customer complaints

and ranking them by urgency. GBM excels at

processing large datasets and modeling intricate

patterns, making it a strong candidate for prioritizing

and classifying customer issues. The model

demonstrated high performance, achieving a

classification accuracy of 90.45% and a prioritization

INCOFT 2025 - International Conference on Futuristic Technology

694

accuracy of 98.84%, particularly excelling in

identifying "urgent" and "not urgent" complaints.

3.3.7 XG Boosting

XG Boost is an advanced gradient boosting algorithm

known for its speed and performance in classification

tasks, making it well-suited for customer complaint

classification and prioritization. By leveraging

decision trees in an ensemble framework, XG Boost

efficiently handles large datasets with missing values

and complex relationships. In the classification task,

the model achieved an accuracy of 91.17% in

categorizing complaints, while for prioritization, it

achieved an outstanding 99.73% accuracy, excelling

at distinguishing between "urgent" and "not urgent"

complaints. Its high precision and recall reflect its

strong predictive performance.

3.4 Deep Learning Models

3.4.1 BERT

BERT (Bidirectional Encoder Representations from

Transformers), is an advanced transformer leveraged

deep learning model developed for natural language

understanding. By employing a bidirectional

attention mechanism, it analyzes a word's context by

considering the words on both sides within a

sentence, making it exceptionally efficient for tasks

like text classification. This makes BERT particularly

suitable for tasks like department categorization and

urgency prioritization, where understanding the

nuanced meaning of words in context is crucial. For

department categorization, BERT achieved an

accuracy of 81%, while for urgency prioritization, it

performed even better with an accuracy of 92%,

showcasing its strong capability in both tasks.

3.4.2 RNN

A Recurrent Neural Network (RNN) is a model

tailored for sequential data processing, using a hidden

state to retain information from prior inputs and

capture dependencies over time. This makes RNNs

especially suitable for tasks involving sequences,

such as text classification in department

categorization and urgency prioritization, where the

order of words or phrases can influence the meaning.

In the department categorization task, RNN achieved

an accuracy of 80%, while for urgency prioritization,

it also achieved an accuracy of 92%, showing that it

can handle both tasks effectively.

3.4.3 CNN

A Convolutional Neural Network (CNN) is typically

used for image processing but has proven effective

for text classification tasks by applying convolutional

layers to extract local patterns from sequences of text.

This ability to detect local features makes CNNs

suitable for department categorization and urgency

prioritization, where the focus is on identifying

important keywords or phrases within text. CNNs are

particularly effective when the task benefits from

recognizing local patterns and structures in data, such

as words or phrases with specific relevance. In

department categorization, CNN achieved an

accuracy of 80%, while for urgency prioritization, the

model achieved an accuracy of 91%, showing its

ability to handle both tasks with competitive

performance.

4 PERFORMANCE ANALYSIS

This study evaluates model performance using

Precision, Recall, F1-Score, and Accuracy. Precision

indicates the ratio of correctly predicted positive

cases to the total positive predictions, showcasing the

model's reliability in positive classifications. Recall,

assesses the model's effectiveness in detecting all

actual positive instances, calculated as the proportion

of true positives among all actual positives. The F1-

Score, offers a balanced evaluation by accounting for

both false positives and false negatives, making it

particularly valuable for imbalanced datasets. Lastly,

Accuracy measures the percentage of all correct

predictions, encompassing both true positives and

true negatives, providing a comprehensive

performance overview.

Table 1: Comparison of ml models for department

categorization

Model Precision Recall F1

Score

Accuracy

(%)

Logistic

Regression

0.93 0.91 0.92 0.92

SVM 0.92 0.91 0.91 0.92

XG Boost 0.91 0.91 0.91 0.91

Gradient

Boosting

Machines

0.90 0.90 0.90 0.90

Random Forest 0.83 0.77 0.78 0.81

Decision Tree 0.78 0.78 0.78 0.78

Multinomial

Naïve Bayes

0.79 0.64 0.61 0.72

Customer Support Ticket Categorization and Prioritization Using Natural Language Processing

695

Table 2: Comparison of dl models for department

categorization

Model Accuracy(%)

BERT 0.81

RNN 0.80

CNN 0.80

Table 3: Comparison of dl models for urgency

prioritization

Model Accuracy(%)

BERT 0.92

RNN 0.92

CNN 0.91

Table 4: Comparison of ml models for urgency

prioritization

Model Precision Recall F1 Score Accuracy

(%)

Decision Tree 1.00 1.00 1.00 1.00

XG Boost 1.00 1.00 1.00 1.00

Gradient Boosting

Machines

0.99 0.95 0.97 0.99

SVM 0.97 0.75 0.82 0.95

Random Forest 0.94 0.53 0.53 0.90

Logistic

Regression

0.93 0.53 0.54 0.90

Multinomial Naïve

Bayes

0.45 0.50 0.47 0.89

5 RESULTS AND DISCUSSION

To find best model for both the categorization of

departmental affiliations and the prioritization of

tickets, the study conducted a thorough evaluation of

the performance of various machine learning and

deep learning architectures. In the department

classification task, logistic regression emerged as the

best-performing machine learning model, achieving

the highest accuracy of 91.99%, closely followed by

SVM (91.51%) and XG Boost (91.17%). Gradient

Boosting Machines demonstrated solid results with

an accuracy of 90.45%, while Random Forest and

Decision Tree performed moderately, achieving

81.13% and 78.44%, respectively. Multinomial

Naive Bayes, however, recorded the lowest accuracy

at 71.87%, indicating its limited suitability for this

task. Among the deep learning models, BERT stood

out with the highest accuracy of 81.00%, showcasing

its ability to leverage pre-trained contextual

embeddings effectively. CNN and RNN followed

closely with accuracies of 80.83% and 80.45%,

respectively, indicating competitive but slightly

lower performance compared to transformer-based

architectures. In the priority classification task,

Decision Tree outperformed all other models,

achieving a remarkable accuracy of 99.96%. Gradient

Boosting Machines (98.84%) and XG Boost

(99.73%) also demonstrated exceptional

performance. Logistic Regression (90.11%), Random

Forest (90.03%), and SVM (94.68%) showed

consistent and reliable results, while Multinomial

Naive Bayes lagged slightly with an accuracy of

89.39%. Among deep learning models, BERT again

proved to be the most effective, achieving an

accuracy of 92.64%, followed closely by CNN

(92.50%) and RNN (91.79%). These results reaffirm

the effectiveness of BERT's pre-trained embeddings

in handling natural language tasks with contextual

complexity.

Figure 2: Department classification ML model’s accuracies

Figure 3: Department classification DL model’s accuracies

0.92 0.92

0.91

0.9

0.81

0.78

0.72

0.2

0.4

0.6

0.8

Accuracy

0.81

0.8 0.8

0.795

0.8

0.805

0.81

0.815

BERT RNN CNN

Accuracy

INCOFT 2025 - International Conference on Futuristic Technology

696

Figure 4: Priority classification ML model’s accuracies

Figure 5: Priority classification DL model’s accuracies

6 CONCLUSIONS

This project represents a significant stride in

automating customer support ticket management,

offering a comprehensive solution for both

department categorization and priority classification.

Logistic regression emerged as a standout performer

in department classification, achieving an impressive

accuracy of 91.99%, closely trailed by SVM and XG

Boost. Conversely, decision tree excelled in priority

classification, boasting an exceptional accuracy of

99.96%. The inclusion of deep learning models,

notably BERT, showcased competitive performance,

particularly in priority classification, achieving an

accuracy of 92.64%. These findings highlight the

efficacy of machine learning (ML) and deep learning

(DL) techniques in enhancing ticket management

processes and elevating customer satisfaction levels.

Deep learning models, like BERT, will exhibit

superior performance when confronted with larger

datasets and complex, unstructured data, as they can

discern intricate patterns and dependencies.

Furthermore, DL models may outperform ML models

when trained over numerous epochs, leveraging their

ability to learn hierarchical representations of data.

Our study extends beyond conventional approaches

by incorporating innovative topic modelling

techniques to label unlabelled data effectively. This

approach not only improves model accuracy but also

demonstrates versatility and practicality in real-world

support environments, addressing the challenge of

limited availability of labelled data when coming to a

real world problem. By integrating both department

categorization and priority classification into a

unified framework, our research achieves remarkable

accuracy levels, surpassing previous studies that

focused solely on one aspect. Moving forward, future

research endeavours may explore additional

optimization strategies and investigate the

deployment of these models in diverse support

environments to validate scalability, effectiveness,

and real-world applicability

REFERENCES

Vairetti, C., Aránguiz, I., Maldonado, S., Karmya, J. P., &

Leal, A. (2024). Analytics-driven complaint

prioritisation via deep literacy and multicriteria

decision-timber. European Journal of Functional

Exploration, 312, 1108–1118.

Aldunate, Á., Maldonado, S., Vairetti, C., & Armelini, G.

(2022). Understanding client satisfaction via deep

literacy and natural language processing. Expert

Systems with Applications, 209, 118309.

Olujimi, P. A., & Ade-Ibijola, A. (2023). NLP ways for

automating responses to client queries: A methodical

review. Published online.

https://doi.org/10.1007/s44163-023-00065-5

Ahmed, H. A., Bawany, N. Z., & Shamsi, J. A. (2021).

CaPBug: A framework for automatic bug

categorization and prioritization using NLP and

machine learning algorithms. IEEE Access,

https://doi.org/10.1109/ACCESS.2021.3069248

Patel, N., & Trivedi, S. (2020). Using predictive modeling,

machine learning personalization, NLP client support,

and AI chatbots to increase client fidelity. EQM

ResearchBerg, 3(3), 1-10.

Ramaswamy, S., & DeClerck, N. (2020). Client perception

analysis using deep learning and NLP. Caterpillar Inc.,

Montgomery, IL 60538.

Qamili, R., Shabani, S., & Schneider, J. (2018). An

intelligent framework for issue ticketing system

grounded on machine literacy. Proceedings of the IEEE

EDOCW 2018, DOI: 10.1109/EDOCW.2018.00022

G, N., Kirthiga, R. V., & Ghayathri, J. (2021). Automatic

trailing of client support tickets using BERT algorithm.

Suhel SF, Shukla VK, Vyas S, Mishra VP, discussion

to robotization in banking through chatbot using

artificial machine intelligence language. In 2020 8th

International Conference on Trustability, Infocom

0.99

0.95

0.9 0.9

0.89

0.8

0.85

0.9

0.95

Accuracy

0.92 0.92

0.91

0.905

0.91

0.915

0.92

0.925

BERT RNN CNN

Accuracy

Customer Support Ticket Categorization and Prioritization Using Natural Language Processing

697

Technologies, and Optimization Trends and Unborn

Directions (ICRITO), IEEE, 611-618.

Blümel, J., & Zaki, M. (2022). Comparative analysis of

classical and deep learning-based natural language

processing for prioritizing customer complaints.

Proceedings of the 55th Hawaii International

Conference on System Sciences, 10 pages.

Silva, S., Pereira, R., & Ribeiro, R. (2018). Machine

learning in incident categorization automation. In 13th

Iberian Conference on Information Systems and

Technologies (CISTI), Caceres, Spain, 13-16 June

2018. IEEE.

Reddy, S. R. S., & Prabhu, V. (2023). Role of Artificial

Intelligence in Request Management: A

Comprehensive Review. In 2023 IEEE 11th Region 10

Humanitarian Technology Conference (R10-HTC).

IEEE.

Lucini, F. R., Tonetto, L. M., Fogliatto, F. S., & Anzanello,

M. J. (2019). Text mining approach to explore

dimensions of airline customer satisfaction using online

customer reviews. Journal of Air Transport

Management, 78, 101760.

Ishizuka, R., Washizaki, H., Tsuda, N., Fukazawa, Y., Ouji,

S., Saito, S., & Iimura, Y. (2022). Categorization and

visualization of issue tickets to support understanding

of implemented features in software development

projects. Applied Sciences, 12(7), 3222.

Al-Hawari, F., & Barham, H. (2019). A machine learning-

based help desk system for IT service management.

Journal of King Saud University-Computer and

Information Sciences.

Benitez Pereira, L. S., Pizzio, R., Bonho, S., Ferraz De

Souza, L. M., & Arnoni Junior, A. C. (2023). Machine

Learning for Classification of IT Support Tickets. In

2023 International Conference on Cyber Management

and Engineering (CyMaEn), Bangkok, Thailand. IEEE.

INCOFT 2025 - International Conference on Futuristic Technology

698