A Multilingual, Context‑Aware E‑Commerce Chatbot Framework for

Personalized Customer Engagement and Real‑Time Sales

Optimization Using Advanced NLP

Abdel Ghaffar Ben Hamida

, J. Sathish Kumar

, Enock I.

, R. Prabha

S. Muthukumar

and Nadimuthu D.

Department of Department of Marketing, College of Business, King Abdulaziz University, Jeddah 21589, Jeddah 21589,

Saudi Arabia

Department of Commerce, Faculty of Science and Humanities, SRM Institute of Science and Technology, Chennai, Tamil

Nadu, India

Department of Management Studies, Nandha Engineering College (Autonomous), Vaikkaalmedu, Erode, Tamil Nadu,

India

Department of Management Studies, Sona College of Technology, Salem, Tamil Nadu, India

Keywords: e‑Commerce Chatbot, Multilingual NLP, Personalized Customer Interaction, Context‑Aware Conversation,

Real‑Time Learning.

Abstract: The development of customer communication in e-commerce is progressing fast with intelligent chatbot

systems based on Natural Language Processing (NLP) being introduced. In this paper, a new multilingual

context-aware chatbot that aims at increasing customer engagement and revenue generation on a variety of e-

commerce websites is presented. In contrast, the proposed model is not only domain independent and dynamic

but gains flexibility in handling the reality-related tasks, both in the code-mixing environment and with real-

time CRM integration and inventory data. Furthermore, the framework has real-time learning abilities that

allow it to learn from changing consumer activities and seasonal trends. Both technical accuracy and user

satisfaction (defined by customer satisfaction, conversions, leads) are evaluation metrics for the chatbot that

shows the scope of better performance results. Blending focus on security and ethical NLP processes to further

push adherence and trust. The system meets the needs of modern e-commerce business looking to expand in

the global market and adopt an intelligent automation system.

1 INTRODUCTION

The rapid rise of e-commerce has changed the way in

which businesses reach their customers, with being

on call, personalized responses, and around the clock

support key to success. As the digital age becomes

even more competitive, traditional models of

customer service have a difficult time keeping up

with the ever-increasing tech-friendly customer who

wants service now, not later. Addressing this seeming

gap, artificial intelligence, particularly Natural

Language Processing (NLP), has emerged as a

promising catalyst to develop receptive human-like

chatbot systems.

Chatbots have come a long way from brittle rule-

based scripts to intelligent conversational agents with

the ability to contextualise and understand user intent.

However, the majority of deployed e-commerce

chatbots are very narrow, with their attention focused

on certain domains, languages or fixed response

flows. Its language-specific constraints make them

less efficient for different searches particularly in a

multi-regional and multilingual exchange

environment. Furthermore, in an age where customer

behaviour is ever-changing, chatbots must be self-

learning beyond existing sets of data so they can adapt

to online requests in real time.

In this paper, we present a new multilingual,

context-aware chatbot framework dedicated to e-

commerce environment for improving the client

engaging and increasing the sales. The platform

utilizes state-of-the-art transformer-based NLP

models, CRM integrations, as well as real-time

Hamida, A. G. B., Kumar, J. S., I., E., Prabha, R., Muthukumar, S. and D., N.

A Multilingual, Context-Aware E-Commerce Chatbot Framework for Personalized Customer Engagement and Real-Time Sales Optimization Using Advanced NLP.

DOI: 10.5220/0013863200004919

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

309-315

ISBN: 978-989-758-777-1

309

learning algorithms, to generate intelligent and

interactive user experiences. Combining technical

prowess with commercial application, the solution

fills the gaps that exist in contemporary chatbots and

architecture, preparing digital commerce now and in

the future.

2 PROBLEM STATEMENT

Although chatbots have been extensively used in e-

commerce platforms, current solutions are unable to

provide intelligent, context-aware and personalized

experiences. Existing solutions are limited by rule-

based domain knowledge, fail to support multi-

lingual and code-mixed content and are unable to

learn dynamic customer behavior in real time. These

constraints lead to inflexible, and hence frequently

inaccurate, interaction: frustration in customers;

lower conversion rates; and, most importantly for this

context, a lack of automation of the sales optimisation

process.

And we can't forget the fact that classical chatbot

architectures rarely plug deeply into systems like

CRM, inventory databases, and recommendation

engines, and so cannot provide the responses you

want (tied to relevant context) or the sales prospecting

you are after. Because of the lack of real-time

adaptation and learning abilities, such systems cannot

be evolved continuously with user feedback and

market development. Moreover, issues around data

privacy, model bias and user experience design are

yet to be fully resolved, thereby posing ethical and

practical challenges on the large-scale deployment of

chatbots.

This paper tries to bridge this gap by proposing a

novel general, scalable and secure NLP driven

chatbot framework that would be multilingual and

context-aware and would also be capable of learning

in real-time trying to revolve the customer

interaction as well as the sales in a futuristic manner

in the e-commerce ecosystem.

3 LITERATURE SURVEY

The development of conversational agents in the e-

commerce is greatly shaped by the recent progress of

Natural Language Processing (NLP), where chatbots

are serving as a key tool to automate customer

assistance. Khennouche et al. (2023) investigate

deployment issues of generative models such as

ChatGPT in FAQ settings, they reveal the

shortcomings of keeping context and domain

relevance. Similarly, Mashaabi et al. (2022)

pioneered a comprehensive review of NLP in

customer service, shedding light on the lack of real-

time adaptability and multilingual understanding.

Kumar and Mishra (2025) and Kanthed (2023)

argue that although most chatbots enhance

accessibility, they do not connect with company’s

end-to-end system like inventory/CRM databases as

a result failing to have an effect on a company’s sales

conversion. Sharma (2025) compared available AI

enabled chatbot frameworks and found usability and

contextual awareness to be significant deficits.

Huseynov (2023) focuses on economic aspect of

chatbots in digital marketing, which demonstrates

that chatbots help to reduce costs while there is a

trade-off on the quality of personalization.

Müller, Schmidt (2024) concentrate on the

reception in Chinese e-commerce of chatbots, where

a more serious concern is that these agents are not

very flexible in their language style, resulting in a less

engaging conversation. Lee, and Park (2024)

emphasize the importance of conversational

commerce and how NLP-based bots convert passive

users to active consumers when context is

maintained. Smith and Johnson (2023) examine sales-

oriented chatbot deployments, but raise objections

about the dearth of user-generated metrics such as

satisfaction and perceived relevance.

Patel & Desai (2025) and Chen & Prentice (2024)

both reinforce the need for chatbots based on

customer persona and behavioural data. Zhang and

Wang (2024) use syntactic parsing methods to

enhance the performance of chatbot, yet semantically

personalized option is still unavailable. Kumar and

Singh (2024) perform comparison of chatbot models

on the basis of precision and recall parameters and

found that fine-tuning has increased the accuracy but

adaptive learning is still missing.

Garcia and Lopez (2024) analyse emotional

concern for chatbot conversation and emphasis

ethical NLP using. References Kim and Lee, 2023

Kim and Lee (2023) Perform user Experience Studies

Find that Fallback and bad context threading

frequently sub-optimize overall satisfaction. Almeida

and Silva’s (2023) future perspective discusses the

contribution of chatbots to open innovation and the

fact that they still struggle to deal with unanticipated

questions due to their rigid nature.

Nguyen, Tran et al. (2024) also study how NLP

advances can benefit smarter Pervasive e-commerce

but they recognize the problem of data bias and small

language coverage. Singh and Gupta (2024) propose

a dynamic chatbot model using current web data for

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

310

better interaction. Brown and Davis (2023) highlight

AI marketing in e-commerce, but expose a gap

between the intelligence of chatbots and marketing

automation tools.

Kumar and Sharma (2024) [20] argue for equal

treatment of NLP through tackling model bias and

Khalid et al. (2024) present emotion-aware chatbots,

but lack multilinguality. Pandya and Holia (2023)

also analyzed deployment of chatbot based on

LangChain not including the assessment in live e-

commerce places. Verloop. io (2025) y Kanishcheva

(2025) hacen hincapié en la necesidad de formación

específica de ámbitos y en el diseño de usuarios.

Lastly, Shirkande et al. (2024) design an e‐

commerce chatbot using keyword−based logic, with

low conversational depth and learning feature.

Throughout this literature, it is clear that there is a

strong necessity for an advanced platform that no

longer solely enables multilinguality and on-the-fly

learning but also heavily integrates with the e-

commerce infrastructure and where ethical and

secure interactions can be performed.

4 METHODOLOGY

The proposed methodology focuses on the design,

development, and deployment of an intelligent

chatbot framework tailored for modern e-commerce

platforms.

Figure 1: Workflow of the proposed multilingual context-

aware e-Commerce chatbot.

Figure 1 implies the Workflow of the Proposed

Multilingual Context-Aware E-Commerce Chatbot.

This system is engineered to offer context-aware,

multilingual interactions while dynamically learning

from user behavior and integrating deeply with

backend systems such as customer relationship

management (CRM), inventory, and recommendation

engines. The methodology is divided into six key

stages: data collection and preprocessing,

multilingual NLP model development, context

management, personalization and integration,

adaptive learning module, and evaluation.

4.1 Data Collection and Preprocessing

To train a high-performing and robust NLP chatbot, a

comprehensive dataset was curated from multiple

sources including e-commerce product descriptions,

real customer chat logs, FAQs, and support tickets

from multilingual platforms. Datasets were cleaned,

anonymized for privacy compliance, and normalized

across English, Hindi, Spanish, and Mandarin.

Preprocessing included tokenization, lemmatization,

and removal of stop words. A translation pipeline

using MarianMT was implemented for multilingual

mapping and alignment. Table 1 gives the dataset

composition across language.

Table 1: Dataset composition across languages.

Language Source Number

Queries

Domain(s)

Covered

English Customer

Chat Logs

12,000 Orders,

Product Info,

Returns

Hindi E-

Commerc

e Forums

8,500 Delivery,

Payments,

Support

Spanish Support

Tickets

7,300 General

Queries, FAQ

Mixed

Input

Code-

mixed

4,000 Offers, Cart

Issues

Figure 2: Distribution of multilingual datasets used in

training the chatbot, showing a balanced composition

across english, hindi, spanish, and code-mixed queries.

Figure 2 illustrates the Distribution of multilingual

datasets used in training the chatbot, showing a

A Multilingual, Context-Aware E-Commerce Chatbot Framework for Personalized Customer Engagement and Real-Time Sales

Optimization Using Advanced NLP

311

balanced composition across English, Hindi, Spanish,

and code-mixed queries.

4.2 NLP Model Development

The core engine of the chatbot uses a fine-tuned

transformer-based architecture. We employed a

hybrid model comprising BERT for intent

classification and a GPT-3.5 (or similar open-source

variant like BLOOM) for response generation. Intent

classification ensures accurate categorization of user

queries, while the generation module provides

coherent, human-like responses. Each model was

fine-tuned using e-commerce-specific datasets to

improve contextual understanding related to

products, orders, payments, returns, and offers.

4.3 Context Management Layer

A context management system was introduced to

support smooth multi-turn conversations with a

LSTM-based memory buffer and attention

mechanisms. This is layer is inspired on the user data

that keeps record on the user’s search history, current

request and shopping behavior and keeps the session

alive in the chatbot for tracking of messages and

keeping the context of the chatbot speech. It allows

the bot to answer follow-up questions or switch to a

different intent mid-conversation correctly.

4.4 Personalization and System

Integration

Profiling of the users and behavioural information is

continuously taken out from the e-commerce

platform through APIs. I'm talking about purchase

history, cart state, browsing behavior, and CRM

information. These elements content into a

recommendation system that is supported by the

collaborative filtering, and the content‐based

filtering algorithms. The chatbot is tailored: it refers

to the user by name and to his/her previous

purchases, and also makes targeted suggestions. It has

close connections to the stock systems, so it can

check for the availability of products, and to the

CRM, so we can see the status of orders and manage

complaints.

4.5 Adaptive Learning Module

The chatbot includes a reinforcement learning

component with feedback loops, enabling it to learn

from successful and failed interactions. A user

satisfaction score is inferred using sentiment analysis

after each session. Based on this feedback, the bot

adjusts its strategy using Q-learning, gradually

improving its response efficiency and tone.

Furthermore, human-in-the-loop supervision is used

to retrain the model periodically with new data.

4.6 Evaluation and Testing

The system is evaluated using both quantitative and

qualitative metrics. Accuracy, precision, recall, and

F1-score are used to evaluate intent detection and

response matching. In addition, user-centric metrics

such as customer satisfaction (via post-chat surveys),

retention rate, average response time, and conversion

rate are used to measure commercial impact. A/B

testing was conducted against a legacy chatbot system

to assess the improvement in performance

The final chatbot was deployed on a mock e-

commerce platform for demonstration purposes and

tested across multiple browsers and devices. Results

from pilot testing demonstrated improved user

satisfaction, reduced handling time, and higher

engagement rates compared to baseline models.

5 RESULT AND DISCUSSION

To evaluate the performance and practical

effectiveness of the proposed multilingual, context-

aware e-commerce chatbot, a comprehensive set of

experiments and user trials were conducted. The

system was deployed on a simulated e-commerce

platform and tested under real-world usage conditions

by a sample group of users interacting in multiple

languages including English, Hindi, and Spanish. The

results are divided into two categories: technical

performance metrics and user engagement insights.

5.1 Technical Performance Analysis

The chatbot’s intent classification model, based on a

fine-tuned BERT architecture, achieved an average

accuracy of 96.4%, with a precision of 95.7%, recall

of 94.9%, and an F1-score of 95.3% across all

supported languages. Compared to a baseline rule-

based model (accuracy: 78.2%), the proposed model

demonstrated a significant improvement in

understanding user queries and correctly identifying

intents across domains like order tracking, product

inquiry, and refund processing. Table 2 gives the

information about performance metrics of NLP

components.

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

312

Table 2: Performance metrics of NLP components.

Component

Accur

Precis

ion

Rec

all

F1-

Score

Intent

Classifie

96.4%

95.7

94.9

95.3%

Response

Generato

– – –

BLEU:

0.82

Multilingual

Detecto

98.1%

97.9

97.5

97.7%

The response generation module, powered by

GPT-3.5 fine-tuned on e-commerce datasets,

exhibited a BLEU score of 0.82, indicating a high

semantic similarity between generated and ideal

responses. Further qualitative testing through blind

user evaluation revealed that 87% of users perceived

the chatbot responses as “natural” or “very natural”,

while only 5% rated them as “robotic.”

Context retention was assessed using multi-turn

conversation tasks. In a session-based evaluation, the

chatbot maintained coherent responses across an

average of 7.2 consecutive turns, compared to 3.4

turns for standard models without contextual

memory. This indicates the effectiveness of the

LSTM-enhanced memory layer in handling complex,

non-linear user conversations.

5.2 Multilingual Capability

Multilingual evaluation was conducted by testing the

chatbot in English, Hindi, and Spanish using both

direct user interaction and synthetically generated

queries. The system achieved comparable

performance across all three languages, with intent

classification accuracy ranging from 95.1% (Hindi) to

96.7% (English). The MarianMT-based translation

pipeline successfully supported code-mixed inputs,

enhancing accessibility for users from multilingual

regions. Table 3 gives the Evaluation result of

languages.

Table 3: Multilingual evaluation results.

Language

Intent

Accuracy

Response

Quality

(

BLEU

)

User

Satisfactio

English 96.7% 0.83 4.7 / 5

Hindi 95.1% 0.81 4.5 / 5

Spanish 95.6% 0.80 4.6 / 5

Furthermore, language-specific colloquialisms

and informal expressions were correctly interpreted

in over 90% of test cases, demonstrating the chatbot’s

robustness in practical multilingual environments.

5.3 User Experience and

Personalization Impact

User engagement metrics collected during the pilot

deployment showed compelling results. The average

session duration increased from 2.3 minutes

(baseline) to 4.9 minutes with the new chatbot,

indicating enhanced user interaction and engagement.

The query resolution rate reached 92.6%, up from

74.5% using the previous system, significantly

reducing the need for human agent intervention.

On a business level, the chatbot led to a 17.4%

increase in conversion rates, especially when adding

personalized product recommendations into the chat.

In addition, there was a 12% decrease in abandoned

carts by means of nudging in real time and

promotional triggers from the context. These results

validate the proactive buying nature of the chatbot as

shown in table 4.

Table 4: Comparison with baseline chatbot.

Metric

Baseline

Chatbot

Proposed

Chatbot

Query Resolution

Rate

74.5% 92.6%

Avg. Response Time

(secs)

4.8 2.1

Fallback Frequency

(%)

15.2% 4.3%

Customer

Satisfaction Score

3.2 / 5 4.6 / 5

5.4 Real-Time Learning and

Adaptation

The reinforcement learning component allowed the

chatbot to learn from user responses. In a two-week

feedback loop, the chatbot strengthened its ability to

deal with ambiguous queries by 8.2%, as measured

by a reduction in fallback instances. Sentiment-based

feedback gathering also indicated that positive user

sentiment was a total of increased by 19.3% when

conversational improvements were made using our

adaptive learning as shown in table 5.

A Multilingual, Context-Aware E-Commerce Chatbot Framework for Personalized Customer Engagement and Real-Time Sales

Optimization Using Advanced NLP

313

Table 5: Adaptive learning improvements over time.

Week

Fallback

Rate (%)

User

Sentiment

(+ve %)

Accuracy

Increase

(%)

1 9.1 63.4 –

2 7.3 71.5 +2.4

3 6.2 78.9 +4.1

4 4.3 82.7 +5.8

5.5 Security and Ethical NLP

Compliance

The chatbot was also evaluated for privacy and

ethical concerns. No PII was retained without

encryption, and all user data logs were anonymized.

The system satisfied GDPR data access, opt-out, and

session tracking transparency criteria. Of note, the

bias detection analysis revealed no statistically

significant bias tendencies with respect to any of the

demographic variables.

6 CONCLUSIONS

Intelligent chatbots have revolutionized customer

service and engagement in digital commerce. This

work presented a new kind of multilingual, context-

aware chatbot for e-commerce platforms with state-

of-the-arts NLP techniques as to perform real-time

learning, as well as hyper-personalisation. Unlike

traditional systems, whose responses are constrained

by static patterns- or language-specific knowledge,

we show that our approach leads to dramatic

improvements in conversational quality, user

satisfaction and downstream commercial metrics.

Using transformer-based models, context tracking

architectures and reinforcement learning along

feedback loops, system provides fluid humaoid

conversations while keeping the coherence over for

multiple turns of conversation. Its multilingualism,

experimented over a variety of languages and c o-

demanded queries, makes it a scalable and universal

solution for the worldwide e-commerce companies.

In addition, the chatbot integration with back-end

systems like CRM, inventory databases, and

recommendation engines creates a buzzworthy

shopping experience that is dynamic, personalized,

and not only solves customer inquiries but proactively

drives sales and user engagement. Its measurements

of privacy give us confidence of its real-world

suitability and its ethically aware NLP practices

make NMN ready for deployment.

Finally, the presented chatbot framework is

anticipated to become a progressive milestone in e-

commerce automation and overcomes efficiency

issues from both scalability, personalization and

intelligence perspectives. It reinvents conversational

commerce, offering companies a unique opportunity

to drive richer customer conversation, streamline

operations and shape the future of commerce in the

digital space.

REFERENCES

Almeida, J., & Silva, T. (2023). Implementation of chatbot

in online commerce and open innovation. Journal of

Open Innovation: Technology, Market, and

Complexity, 9(4), 894. https://doi.org/10.1016/j.joi.20

22.100894

Brown, T., & Davis, L. (2023). The impact of artificial

intelligence marketing on e-commerce sales. Systems,

12(10), 429. https://doi.org/10.3390/systems12100429

MDPI

Chen, Y., & Prentice, C. (2024). Integrating artificial

intelligence and customer experience. Australasian

Marketing Journal, 32(1), 45– 56. https://doi.org/10.10

16/j.ausmj.2024.01.005

Garcia, M., & Lopez, R. (2024). Serving customers through

chatbots: Positive and negative effects on customer

experience. Journal of Service Theory and Practice,

34(2), 123–140. https://doi.org/10.1108/JSTP-01-

2023-0015

Huseynov, F. (2023). Chatbots in digital marketing:

Enhanced customer experience and reduced customer

service costs. In Digital Marketing Strategies (pp. 47–

65). IGI Global. https://www.researchgate.net/publicat

ion/372837440

Kanthed, S. (2023). The role of chatbots in reducing

customer support response time in e-commerce.

International Journal of Scientific Research in Enginee

ring and Management, 7(12), 1– 7. https://www.resear

chgate.net/publication/390124084

Khennouche, F., Elmir, Y., Djebari, N., Himeur, Y., &

Amira, A. (2023). Revolutionizing customer interactio

ns: Insights and challenges in deploying ChatGPT and

generative chatbots for FAQs. arXiv. https://arxiv.org/

abs/2311.09976

Kim, S., & Lee, D. (2023). Understanding the user

experience of customer service chatbots. International

Journal of Human-Computer Studies, 159, 102738.

https://doi.org/10.1016/j.ijhcs.2022.102738

Kumar, A., & Mishra, S. (2025). Role of AI-chatbots in

enhancing e-commerce accessibility and its impact on

customer experience. Asian Journal of Management

and Commerce, 6(1), 729– 735. https://doi.org/10.222

71/27084515.2025.v6.i1h.520

Kumar, N., & Singh, P. (2024). Advanced NLP models for

technical university information chatbots: Developme

nt and comparative analysis. International Journal of

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

COMMUNICATION, AND COMPUTING TECHNOLOGIES

314

Advanced Computer Science and Applications, 15(3),

112– 119. https://www.researchgate.net/publication/37

8352411

Lee, H., & Park, J. (2024). AI-based chatbots in

conversational commerce and their effects on customer

engagement. Journal of Retailing and Consumer

Services, 67, 102063. https://doi.org/10.1016/j.jretcons

er.2021.102063

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

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

customer service: A systematic review. arXiv. https://a

rxiv.org/abs/2212.09523

Müller, J., & Schmidt, L. (2024). Understanding user

acceptance of AI-driven chatbots in China's e-

commerce sector. Systems, 13(2), 71. https://doi.org/1

0.3390/systems13020071

Nguyen, T., & Tran, H. (2024). Advancements in natural

language processing: Implications for e-commerce

chatbots. Journal of Artificial Intelligence Research, 77,

59–78. https://doi.org/10.1016/j.jair.2024.03.005

Patel, R., & Desai, M. (2025). Enhancing customer service

effectiveness with AI-powered chatbots. International

Journal of Research Publication and Reviews, 6(4),

1369– 1375. https://ijrpr.com/uploads/V6ISSUE4/IJR

PR41369.pdf

Sharma, R. (2025). AI-powered chatbot for customer

support: A review of recent advancements. Internation

al Research Journal of Modern Engineering and

Technology Science, 3(3), 45– 52. https://www.irjmets

.com/uploadedfiles/paper/issue_3_march_2025/70205/

final/fin_irjmets1744474775.pdf

Singh, A., & Gupta, R. (2024). Dynamic e-commerce

website with NLP chatbot: Enhancing user interaction.

International Journal of Computer Applications,

182(7), 25– 30. https://www.researchgate.net/publicati

on/388822052

Smith, A., & Johnson, B. (2023). The impact of AI-

powered chatbots on customer service and e-commerce

sales. European Economic Letters, 12(4), 2392–2405.

https://www.eelet.org.uk/index.php/journal/article/do

wnload/2392/2147/2645

Zhang, L., & Wang, Y. (2024). Optimizing chatbot

effectiveness through advanced syntactic analysis.

Applied Sciences, 14(5), 1737. https://doi.org/10.3390

/app14051737

A Multilingual, Context-Aware E-Commerce Chatbot Framework for Personalized Customer Engagement and Real-Time Sales

Optimization Using Advanced NLP

315