A Scalable, Multilingual and Emotion‑Aware NLP Framework for

Intelligent and Secure Automated Customer Support Systems

Ramakrishna Kosuri

, K. Prabakar

, M. Elakkiya

, R. Prabha

, V. Sukash

and A. A. Nagamani

Engagement Manager, Tata Consultancy Services, Computer consultant, Celina, Texas, 75009, U.S.A.

Department of MBA, Hindusthan College of Arts & Science, Behind Nava India, Coimbatore - 641 028, Tamil Nadu, India

Department of MBA, J.J. College of Engineering and Technology, Tiruchirappalli, Tamil Nadu, India

Department of Management Studies, Nandha Engineering College (Autonomous), Erode‑Perundurai Main Road,

Vaikkaalmedu, Erode District‑638052, Tamil Nadu, India

Department of Computer Science and Engineering MLR Institute of Technology, Hyderabad, Telangana, India

Keywords: NLP, Customer Support Automation, Multilingual Chatbot, Emotion‑Aware AI, Secure Conversational

Systems.

Abstract: With the advent of digital communication, the proliferation of customer support becomes insatiable. We

herein develop an emotion-aware, scalable, multilingual natural language processing (NLP) solution for

automatised reactivity of customer service user interactions. The system combines cutting-edge transformer

models, sentiment tracking, and culture sensitivity layers in order to provide personalized and emotionally

intelligent responses in a number of platform, such as voice and text interfaces. It exploits transfer learning

and online KB update for adaptation with strong guarantees of data privacy and scalable solutions with cloud-

based implementation. They conduct an extensive evaluation with real-world dataset including accuracy as

well as customer satisfaction with retention and resolution efficiency. The model is also smart to have

escalation built in for if humans need to step in. The findings show significant enhancements in user

satisfaction, operational efficacy, and long-term ability to adapt over traditional customer support technology.

1 INTRODUCTION

This is especially true in the digital age, where

customer care is no longer defined by earphone

touting call center agents but as a part of the online

service universe. They demand instant, precise, and

custom responses—no matter the channel, time or

language. As customer bases scale and expectations

soar, businesses are leveraging Artificial Intelligence

(AI) – and particularly Natural Language Processing

(NLP) – to automate and improve the customer

experience.

However, even if NLP became common in

chatbots, although most of the current solutions have

important limitations, such as operating

monolingualy, low emotional intelligence,

nonscalability of scripted answer, etc. These have led

to less user satisfaction and a higher risk of customer

churn. But then that’s the problem with data-driven

approaches: they are always behind the curve, and

coming off them risks a harsh withdrawal. But then

again that's the thing about data-based approaches:

they're always playing catch up, and going cold

turkey is no picnic. And then there's the issue of how

to understand other cultures, other languages, other

idioms and other expressions.

To fill these gaps, this work presents a scalable,

multilingual, and emotion-aware NLP-based

customer support framework. The featured system

adopts the state-of the art deep learning methods, real-

time sentiment analysis, and cultural sensitivity

modeling, to follow on the behavior of user’s request

in an intelligent and sensitive manner. Unlike

traditional chatbots, this system is designed to learn

overtime through user response and database updates,

and as such, it continues to improve and become

more effective.

The use of transfer learning from pre-trained

transformer models (e.g. BERT, GPT) and secure

cloud-based deployment, makes the system resource

efficient and easily scalable. Furthermore, our

method can be able to hand off complex or

Kosuri, R., Prabakar, K., Elakkiya, M., Prabha, R., Sukash, V. and Nagamani, A. A.

A Scalable, Multilingual and Emotion-Aware NLP Framework for Intelligent and Secure Automated Customer Support Systems.

DOI: 10.5220/0013863000004919

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

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 1, pages

301-308

ISBN: 978-989-758-777-1

301

incomprehensible requests to human agents,

maintaining a human-in-theloop mechanism

whenever needed.

We show how this system can be

ARCHITECTED and EVALUATED for real-world

datasets and use-cases, significantly improving

resolution time, user experience, and system

flexibility. In solving for the fundamental flaws of

legacy systems, this framework is becoming the new

benchmark for intelligent, responsive, secure

automated customer support.

2 PROBLEM STATEMENT

Challenges with Traditional Automated Customer

Support Traditional automatic customer support

systems are frequently unable to provide genuinely

effective and human-like support for people due to a

number of inherent limitations. Many of the current

models are limited to monolingual interactions, do

not have emotional understanding, and do not scale

well across high number of users or platforms. They

use static rule-based responses that are insensitive to

the dynamically changing user intents and

sentiments or culture sensitivity. Further, they do not

address satisfactorily the issues along the lines of user

privacy, real-time update of the user knowledge, and

smooth transition to a human agent. These limitations

lead to impersonal exchanges, user frustration, and

suboptimal service. Hence there is a significant

demand for a scalable and multilingually accessible,

NLP-enabled solution that is both emotionally aware

as well as capable of enhancing user experience as

well as delivering safe and contextually intelligent

support automation for the customers.

3 LITERATURE SURVEY

The pervasive integration of NLP into service

automation has seen several significant

improvements in recent years, mainly due to the rise

of transformer-based models. In lets get to that

already, Arif men et al (2024) described the essential

features of NLP autonomous support system and

pointed out that they can effectively serve faster and

cheaper while maintaining high quality guidance to

the customers. But their experiments were just

theoretical and their system deployment was not

verified. Oladele et al. (2025) further more expanded

it by centering the integrations of NLP in fintech

chatbots, however their research was limited in

financial sectors and only support English.

The transformation of rule-based to intelligent

deep learning models is effectively presented in

Shiva et al. (2024) proposed a humanizing language

model based on the transformer architecture and a

chatbot with no personalized and emotional

functions. Sapanakolambe et al. (2024) suggested

automatic query management but did not illustrate

how the system would scale with multiple languages.

Mashaabi et al. Couto (2022) performed a systematic

review of NLP in customer service and found that

almost all the models have poor contextual

consistency and difficulties with ambiguous

utterances.

Chaidrata et al. (2022) proposed an intent-

matching model to enhance the relevance of the

response, but did not contain adaptive sentiment

analysis. Meanwhile, He et al. (2022) and Salcedo

Gallo et al., (2022) addressed sentiment-aware NLP

systems to detect losing agents in support chats, as a

precursor to emotion-sensitive response generation.

However, scaling and cultural adaptation of these

models was problematic. Niederwieser et al. (2025)

introduced a pragmatic NLP automation model, but

provided little evidence of scalability, security or

multilingual support.

Chatbots as the ones mentioned in Brush and

Zúquete (2022) and Gallo et al. (2022) proposed

advancements in natural response generation, without

addressing long-term context remembering, or

escalation protocols. Katragadda (2023) reviewed

NLP-driven chatbots made more accurate but

multilinguistically gobsmacked and cross-platform

inconsistent. Alotaibi et al., 2022 focused on emotion

recognition and did not include voice support or

feedback loops in their system.

In addition, considering the works of Nwokedi &

Nwafor (2024) and Thakkar et al. (2024) highlighted

the absence of feedback and real-time learning in

current model and explicit evidence accumulation for

decision making. Subsequent research (ScienceDirect

2024; SpringerLink 2023) has proposed the use of

large language models (LLMs), though cost-

optimized training pipelines or bias mitigating LLMs

would be required. Others also, such as SSRN (2025),

emphasized the NLP potential in self-service

applications with reduced support, but had no

escalation capability.As a whole, the literature shows

that despite the application of NLP, the automated

customer service suffers from notable gaps:

Multilingual assistance, Emotional support, Long-

term adaptability, Data security, and Synchronization

with human agents. These observations motivate the

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

302

current work to work on a more complete, scalable,

and secure NLP-based framework for customer

support that can work around these limitations.

4 METHODOLOGY

Motivation em-v2.0 introduces an open-source

multilingual customer support system enhanced with

sentiment-specific sentences which ustomers can

query, insert captivating terms w.o.arguably being a

party to sth. With a combination of state-of-the-art

deep learning techniques and rock-solid architecture,

emotion, and real-time adaptability, the system has

been built to provide the most natural and intuitive

response possible. The methodology encompasses six

main steps: data collection and pre- processing,

multilingual embedding fusion, intent classification

and sentiment analysis, contextual response

generation, dynamic escalation resolution, and

system evaluation.

4.1 Data Collection and Preprocessing

Step 1: Building the question corpus Starting with the

initial collection of the 2003 customer support

questions and other question sets available in publicly

accessible sources, and also collecting the domain

specific questions for the 10C challenge day in travel

domain. Sources also comprise multi-turn customers

support conversations from eg Kaggle,

conversational datasets from customer service in

several languages, synthetic data by using data

augmentation methods. “Table 1 presents datasets

used to train and test, which span a variety of

languages and sentiment classes.” The pre-processing

on text data includes standard operations like

lowercasing, punctuation removal, tokenization, stop

word removal and lemmatization. Language

detection is also used to classify inputs for

multilingual routing.

Table 1: Dataset Description for Multilingual and Sentiment-Aware Training.

Dataset Name Language(s) Query Count Sentiment Labels Source

CustomerChatQA English 10,000

Positive, Neutral,

Negative

Kaggle

MultilingualSupport-

100

Spanish, French, Hindi 8,500 Frustrated, Satisfied Open Source

RetailAssist-NLP English, Tamil 7,200 Confused, Angry, Happy Proprietary

CallCenterLogs English 5,000 Neutral, Angry Web-scraped

SyntheticMixGen Multilingual 6,000 All above Augmented

4.2 Multilingual Embedding and

Transformer Integration

To enable support across various languages, we

integrate multilingual BERT (mBERT) and XLM-

RoBERTa embeddings that support over 100

languages. These pre-trained models are fine-tuned

using domain-specific queries to improve context

relevance and intent recognition. A hybrid

transformer-based architecture is used to manageboth

short and long conversation threads, with positional

encodings and attention mechanisms enabling

context retention over multiple dialog turns.

4.3 Intent Classification and Sentiment

Detection

A dual-stream classification model is trained to

identify user intent and underlying sentiment in each

query. The intent classifier categorizes inputs into

predefined service domains (e.g., billing, technical

support, returns), while the sentiment classifier

assesses emotional tone using an LSTM-enhanced

attention layer. This helps prioritize emotional or

urgent queries and ensures tone-appropriate

responses. Emotion labels such as “frustrated,”

“neutral,” “satisfied,” or “confused” are used to

modulate the language and structure of the reply.

4.4 Contextual Response Generation

and Knowledge Base Integration

Instead of fixed rule-based replies, the model

generates context-aware responses using a fine-tuned

T5 or GPT-based model, depending on the

deployment constraints. These responses are enriched

using a dynamic knowledge base that is regularly

updated through APIs connected to FAQ systems,

product databases, and CRM systems. The generated

A Scalable, Multilingual and Emotion-Aware NLP Framework for Intelligent and Secure Automated Customer Support Systems

303

reply also considers previous conversation turns, user

profile data, and detected sentiment to deliver

personalized and accurate answers.

4.5 Escalation Mechanism and

Feedback Loop

For queries deemed too complex, emotionally

escalated, or unresolved after two iterations, the

system triggers an automated escalation to a human

agent. This handover includes conversation context,

identified intent, and sentiment score to minimize

redundancy. Simultaneously, a feedback mechanism

captures user ratings and sentiment post-interaction.

This data is used to retrain components of the model

in an online learning setup, improving system

adaptability over time.

4.6 Deployment and Evaluation

The system is deployed on a cloud-based platform

(e.g., AWS or GCP) using Docker and Kubernetes for

scalable microservices. REST APIs handle

interactions across text and voice channels, with the

voice interface using Google Dialogflow or similar

NLP APIs. Evaluation metrics include intent

accuracy, response latency, sentiment detection

precision, user satisfaction score, average resolution

time, and escalation frequency. These are

benchmarked against traditional rule-based systems

and basic chatbot models using A/B testing in real-

world scenarios. Figure 1 shows the

Workflow of the

Proposed Emotion-Aware Multilingual NLP Support

System.

Figure 1: Workflow of the Proposed Emotion-Aware

Multilingual NLP Support System.

By integrating multilingual NLP, emotion-aware

modeling, real-time updates, and seamless human

handoff, this methodology offers a complete and

adaptive solution for next-generation automated

customer support systems.

5 RESULTS AND DISCUSSION

To evaluate the effectiveness of the proposed

scalable, multilingual, and emotion-aware NLP

framework for automated customer support, a series

of experiments were conducted using real-world

datasets and simulated customer interactions across

multilingual platforms. The evaluation focused on

multiple dimensions: system accuracy, user

satisfaction, latency, emotional responsiveness, and

scalability. These performance metrics were

benchmarked against two baselines: a traditional rule-

based chatbot and a basic transformer-based chatbot

without emotional and multilingual support.

5.1 Intent Recognition and

Multilingual Understanding

The model achieved an intent classification

accuracy of 94.2%, outperforming the rule-based

baseline (81.5%) and the basic transformer model

(88.9%). The use of multilingual embeddings from

XLM-RoBERTa and mBERT significantly enhanced

the system’s ability to interpret queries in over 20

languages. Precision and recall remained consistently

above 90% even in code-switched scenarios,

confirming the model's robustness in multilingual and

regional environments.

User queries in Hindi-English, Spanish-English, and

Tamil-English hybrids were tested and correctly

classified in more than 92% of the cases. These

results highlight the system’s ability to scale across

diverse user bases without requiring separate

language-specific models.

5.2 Sentiment Detection and Emotional

Awareness

The sentiment classification module exhibited a

macro F1-score of 91.3% in detecting emotional

cues such as frustration, satisfaction, confusion, and

urgency. The incorporation of an attention-based

LSTM over contextual embeddings helped the model

dynamically adjust the tone of its responses. As

shown in Table 2, our framework significantly

outperforms rule-based and transformer-only models

in accuracy, response time, and escalation handling.”

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

304

For instance, frustrated users received more

empathetic and solution-oriented replies, while

satisfied users were greeted with appreciation

messages or upsell prompts. Figure 2 show the intent

classification accuracy comparison.

Figure 2: Intent Classification Accuracy Comparison.

Table 2: Performance Comparison of NLP Models Used.

Model

Intent

Accur

acy

(%)

Sentime

nt F1-

score

(%)

Avg.

Respon

se Time

(s)

Escal

ation

Rate

(%)

Rule-Based

Chatbot

81.5 64.2 28.9 13.9

Transformer

(no

emotion/mu

ltilingual)

88.9 78.4 22.1 9.7

Proposed

Framework

94.2 91.3 13.7 5.2

Table 3: Emotion Classification Results Across Languages.

Language Precision (%) Recall (%)

F1-score

(%)

English 92.5 90.7 91.6

Hindi 89.2 88.0 88.6

Spanish 91.0 89.5 90.2

Tamil 88.7 87.9 88.3

French 90.3 89.0 89.6

Compared to the baseline models, which lacked any

emotional intelligence, user engagement increased by

27% when the emotional tone was matched

appropriately. “Table 3 demonstrates the model's

robustness in detecting user sentiment across

different languages.” This validates the hypothesis

that emotionally aware systems foster a more human-

like and effective customer experience. Figure 3

shows the across language graph.

Figure 3: Sentiment Detection F1-Score Across Languages.

5.3 Response Quality and Contextual

Relevance

One of the critical goals was to generate contextual

responses that adapt over multi-turn conversations.

Using a fine-tuned GPT-2 model, responses

demonstrated high fluency and context preservation,

maintaining coherence over five or more turns of

dialogue. In comparison, the baseline models often

failed to retain user context beyond two turns, leading

to repetitive or irrelevant responses.

Manual evaluation by customer support agents

rated the system's responses as “excellent” or “very

good” in 88.5% of cases, whereas the baseline models

received favorable ratings in only 64.3% and 75.1%

of cases, respectively.

5.4 Resolution Time and Escalation

Efficiency

The average query resolution time was reduced to

13.7 seconds, compared to 28.9 seconds and 22.1

seconds for the rule-based and standard transformer

models. The inclusion of dynamic knowledge base

integration and intent-sentiment dual tracking

minimized redundant query loops. Additionally, the

intelligent escalation module accurately flagged and

transferred unresolved or high-sentiment queries to

human agents in just 5.2% of total interactions,

significantly lower than the 13.9% escalation rate

observed in baseline systems.

Escalation handovers were enhanced by

transferring a structured context package—including

user intent, sentiment trajectory, and dialogue

history—thereby enabling human agents to resolve

queries in 32% less time on average.

A Scalable, Multilingual and Emotion-Aware NLP Framework for Intelligent and Secure Automated Customer Support Systems

305

5.5 User Satisfaction and Feedback

Analysis

User satisfaction was assessed via post-interaction

surveys and sentiment analysis of feedback logs. The

overall user satisfaction and feedback breakdown are

presented in Table 4 The proposed system attained a

user satisfaction score of 4.6/5, in contrast to 3.7/5

and 4.1/5 for the rule-based and basic transformer

counterparts. Users appreciated the platform’s

fluency, emotional understanding, and multilingual

capabilities.

Table 4: User Feedback Distribution After Interaction.

Feedback Type Count Percentage (%)

Very Satisfied 1,420 47.3

Satisfied 1,010 33.6

Neutral 370 12.3

Unsatisfied 130 4.3

Very Unsatisfied 70 2.5

Feedback loop analysis showed that 76% of the

suggestions provided through user ratings were

automatically integrated into model retraining via the

online learning pipeline. This iterative improvement

led to noticeable performance gains over time and

demonstrated the model’s capacity for continuous

self-enhancement. Figure 4 shows the user feedback

distribution.

5.6 Scalability and Latency

Performance

Scalability tests were conducted on AWS using a

Kubernetes-based deployment, simulating up to

10,000 concurrent user sessions. System performance

Figure 4: User Feedback Distribution.

under increasing user loads is summarized in Table 5.

The model maintained a 95% response latency under

1.5 seconds, indicating strong potential for enterprise-

scale deployment. Load balancing and microservice

architecture ensured optimal memory usage and

computational efficiency. Table 5 shows the system

performance.

Table 5: System Scalability Performance Under Load.

Concurrent

Users

Avg.

Response

Time (s)

Max

Memory

Usage (MB)

Through

put

(queries/

sec)

100 0.8 650 70

1,000 1.1 1,420 640

5,000 1.3 2,900 2,500

10,000 1.5 3,800 4,800

Figure 5: Average Response Time vs Concurrent Users.

Unlike monolithic chatbot designs, the proposed

system scaled horizontally with minimal increase in

resource consumption, making it suitable for diverse

business domains including e-commerce, healthcare,

and fintech. Figure 5 shows the average response time

and concurrent users.

6 CONCLUSIONS

This work introduces a scalable, multilingual, and

emotion-aware NLP pipeline, and establishes a more

general, future-oriented perspective on automated

customer support. Compared to traditional systems

that are rule-based and focused on language

limitations, the model intelligently handles user

questions, dynamically adjusts emotional tone, and

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

306

responds in context in a personalized way. The

combination of transformer-based models, real-time

sentiment analysis, multilingual embeddings, and

adaptive knowledge bases has resulted in an overall

great improvement in user engagement as well as

support efficiency.

Its multilingual support, ability to retain multi-

turn context and to escalate unresolved problems to

human agents make it scalable and applicable in real-

world settings. Focus testing has shown significant

increases in the important metrics of intent

recognition accuracy, latency of response, user

satisfaction and emotional sensitivity.

(4) Continuous Learning In addition, the design

supports continued learning and advancing by means

of feedback loops, gradually making the model

smarter and more user-centered after each interaction.

Built with security, privacy compliance and cloud

scale in mind, the framework stakes its claim as a

potential solution for companies looking to augment

digital customer service with intelligence and

empathy.

Put simply, this study is a contribution to the

bridging of the technical-automation interface with

human-like customer understanding, towards

intelligent service ecosystems. Ongoing forays into

behavioral analytics, and voice-enabled capabilities,

and industry-specific tuning will solidify its place in

the changing terrain of AI-empowered customer

support.

REFERENCES

Advancements in natural language processing: Implications

for customer service. (2024). ScienceDirect.

https://www.sciencedirect.com/science/article/pii/S27

72503024000598ScienceDirect

Arif, W., & [Co-author]. (2024). Leveraging natural

language processing (NLP) for automated customer

support systems. ResearchGate. https://www.research

gate.net/publication/384678656_Leveraging_Natural_

Language_Processing_NLP_for_Automated_Custome

r_Support_SystemsResearchGate

Chaidrata, A., Shafeeu, M. I., Chew, S. K., Chen, Z., Cham,

J. S., Yong, Z. L., & Yap, U. H. (2022). Intent

matching-based customer services chatbot with natural

language understanding. arXiv preprint

arXiv:2202.00480. https://arxiv.org/abs/2202.00480

arXiv

Conversational natural language processing for automated

customer support services. (2024). ResearchGate.

https://www.researchgate.net/publication/370151386_

Conversational_natural_language_processing_for_aut

omated_customer_support_servicesResearchGate

Enhancing customer service automation and user

satisfaction through AI-powered chatbots. (2023). The

Science Brigade. https://thesciencebrigade.com/jcir/ar

ticle/view/249The Science Brigade+1AIML Studies+1

He, K., Wang, J., Sun, C., & Wu, W. (2022). Unified

knowledge prompt pre-training for customer service

dialogues. arXiv preprint arXiv:2208.14652.

https://arxiv.org/abs/2208.14652arXiv

Katragadda, V. (2023). Automating customer support: A

study on the efficacy of machine learning-driven

chatbots and virtual assistants. IRE Journals, 7(1), 600–

608. https://www.irejournals.com/formatedpaper/170

48601.pdfIre Journals

Leveraging natural language processing in self-service BI

for non-technical users. (2025). SSRN.

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5

059060SSRN

Mashaabi, M., Alotaibi, A., Qudaih, H., Alnashwan, R., &

Al-Khalifa, H. (2022). Natural language processing in

customer service: A systematic review. arXiv preprint

arXiv:2212.09523. https://arxiv.org/abs/2212.09523

arXiv

Natural language processing in customer support. (2024).

International Research Journal of Innovations in

Engineering and Technology, 8(11), 2446–2455.

https://irjiet.com/Volume-8/Issue-11-November-

2024/Natural-Language-Processing-in-Customer-

Support/2446IRJIET

Natural language processing (NLP) in chatbot customer

service. (2025). ResearchGate. https://www.researchg

ate.net/publication/390337559_Natural_Language_Pro

cessing_NLP_in_Chatbot_Customer_Service

ResearchGate

Niederwieser, E., Siegele, D., & Matt, D. T. (2025).

Automating customer service using natural language

processing. ResearchGate. https://www.researchgate.

net/publication/383161189_Automating_Customer_Se

rvice_Using_Natural_Language_Processing

ResearchGate

NLP techniques for automating responses to customer

queries: A systematic review. (2023). SpringerLink.

https://link.springer.com/article/10.1007/s44163-023-

00065-5SpringerLink

Nwokedi, C. C., & Nwafor, C. A. (2024). Enhancing

customer service and user experience through the use of

machine learning-powered intelligent chatbots. World

Journal of Advanced Research and Reviews, 14(2),

2307– 2315. https://wjarr.com/sites/default/files/WJA

RR-2024-2307.pdfWjarr

Oladele, S., Favour, A., Kunle, A., & Timileyin, J. (2025).

The role of natural language processing (NLP) in

enhancing customer experience in fintech: Applications

in chatbots and virtual assistants. ResearchGate.

https://www.researchgate.net/publication/390454097_

The_Role_of_Natural_Language_Processing_NLP_in

_Enhancing_Customer_Experience_in_Fintech_Appli

cations_in_Chatbots_and_Virtual_Assistants

ResearchGate

Salcedo Gallo, J. S., Solano, J., García, J. H., Zarruk-

Valencia, D., & Correa-Bahnsen, A. (2022). Proactive

A Scalable, Multilingual and Emotion-Aware NLP Framework for Intelligent and Secure Automated Customer Support Systems

307

detractor detection framework based on message-wise

sentiment analysis over customer support interactions.

arXiv preprint arXiv:2211.03923. https://arxiv.org/abs

/2211.03923arXiv

Sapanakolambe, S., & [Co-authors]. (2024). Enhancing

customer service efficiency: Automating responses to

customer queries using NLP. Nanotechnology

Perceptions, 20(S6), 804–814. https://nano-

ntp.com/index.php/nano/article/download/1371/1149/

2461Nanotechnology Perceptions

Shiva, K., Etikani, P., Bhaskar, V. V. S. R., Kaur, J.,

Thakkar, D., Kanchetti, D., & Munirathnam, R. (2024).

Natural language processing for customer service

chatbots: Enhancing customer experience. International

Journal of Intelligent Systems and Applications in

Engineering, 12(3), 6405–6412.

https://www.ijisae.org/index.php/IJISAE/article/view/

6405IJISAE

Utilizing AI and natural language processing for enhanced

customer service. (2024). Journal of Artificial

Intelligence Research and Applications, 3(1), 314–325.

https://aimlstudies.co.uk/index.php/jaira/article/view/3

14AIML Studies

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

308