Sentiment Analysis-Based Chatbot System to Enhance Customer

Satisfaction in Technical Support Complaints Service for

Telecommunications Companies

Anghelo Juipa

, Luis Guzman

and Edgar Diaz

Peruvian University of Applied Sciences, Antonio Miro Quesada, Lima, Peru

Department of Computing and Informatics, Peruvian University of Applied Sciences, Lima, Peru

Keywords: Chatbot, Natural Language Processing, Emotions, Satisfaction, Customer Service, Sentiment Analysis,

Technical Support Complaints, GPT 3.5.

Abstract: In the competitive world of telecommunications, a good customer technical support complaint service can

make a difference. However, this business process still presents deficiencies in its quality. In the capital of

Peru, there were 102,665 internet complaints and 38,621 cable television complaints. 9.27% and 9.97% of

these, respectively, weren’t resolved. In this sense, this research proposes the implementation of a chatbot,

which incorporates GPT 3.5 as a sentiment analysis component, to reduce user dissatisfaction in this service

process. To validate the proposal, experiments were conducted with 50 internet and cable television service

owners to evaluate satisfaction and accuracy in recognizing their emotions. The results indicated that 86% of

the respondents were satisfied with the chatbot service, and the satisfaction index reached 77.9, surpassing

the minimum threshold of 75 points for providing quality customer service established by the industry. The

methodology behind these results is detailed in the following research.

1 INTRODUCTION

In Peru, customer service quality indices are

deficient. According to the report on service quality

prepared by its Organismo Supervisor de Inversión

Privada en Telecomunicaciones (OSIPTEL, 2021),

none of the telecommunications companies operating

there manage to meet the minimum threshold of 75

points necessary to guarantee quality service. This

deficiency in service quality is clearly reflected in the

complaints recorded during the year 2021 in the

capital of the country, which were related to technical

support, specifically in internet services (102,665)

and cable television (38,621). Of these complaints, a

significant percentage, 9,522 (9.27%) and 3,850

(9.97%) respectively, weren’t satisfactorily resolved

(OSIPTEL, 2022). In response to this situation and

according to a survey conducted by Spiceworks

(2018), chatbots emerged as a support tool to improve

the customer service of companies, thus being the

https://orcid.org/0009-0003-1961-9954

https://orcid.org/0009-0000-6058-3357

https://orcid.org/0000-0003-2101-5503

Customer Service/Customer Support department

(20%) is positioned in third place among the

departments that use chatbots the most in their daily

tasks. This is observed in the research conducted by

Kainathan et al. (2021), who present the chatbot

XiaoIce, which employs anthropomorphic

characteristics aimed at enhancing customer service.

Similarly, Ngai et al. (2021) develop a knowledge-

based chatbot to support customer service in e-

commerce sales and marketing. However, despite the

utility offered by customer service chatbots, current

approaches are limited to specific task resolution or

responding to inquiries using a predefined knowledge

base. Additionally, the emotional dimension of the

user during the interaction between human and

chatbot is not adequately considered, an essential

aspect for understanding and effectively addressing

technical support-related complaints. Therefore, this

research is based on applying the knowledge from

these studies and extending it to the field of

telecommunications, considering user sentiment

Juipa, A., Guzman, L. and Diaz, E.

Sentiment Analysis-Based Chatbot System to Enhance Customer Satisfaction in Technical Support Complaints Service for Telecommunications Companies.

DOI: 10.5220/0012807200003764

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

In Proceedings of the 21st International Conference on Smart Business Technologies (ICSBT 2024), pages 28-36

ISBN: 978-989-758-710-8; ISSN: 2184-772X

analysis to achieve positive outcomes regarding

customer satisfaction.

The key components of the research involve GPT-

3.5, which will be used for recognizing user emotions.

Furthermore, this technology is integrated within the

VoiceFlow platform for the development of the

chatbot, which in turn is integrated with Airtable to

manage customer data.

That said, the main contributions are as follows:

 Implementation of a chatbot system using

GPT-3.5 capable of recognizing user intention

and emotion during interaction thanks to its

sentiment analysis component.

 Application of anthropomorphic features such

as empathy, warmth in conversations, and

humor to enhance complaint handling.

 Validation of the proposed solution through

experiments in a controlled environment

involving 50 individuals who are subscribers of

internet and/or cable television services.

On the other hand, the results of the present

research, obtained after conducting experiments in a

controlled environment with 50 users who are

subscribers of internet and cable television services in

the northern part of Lima, reveal that 86% of the

participants expressed high satisfaction with the

attention provided by the chatbot. Furthermore, 56%

of the users highlighted the chatbot's ability to detect

emotions as very high. In terms of the Customer

Satisfaction Index calculated using the ACSI formula,

a score of 77.9 was recorded, surpassing the

minimum required score for satisfaction according to

industry standards. These findings conclusively

support the effectiveness and acceptance of the

chatbot in improving the customer experience in the

realm of internet and cable services.

This paper is distributed across 5 sections. Section

2 reviews related works on customer service chatbots.

Then, it addresses the concepts and theories related to

the background of this research, as well as the main

contribution we will present in Section 3. Section 4

explains the experiments conducted in a controlled

environment. Finally, this work concludes with the

presentation of the conclusions derived from this

project.

2 RELATED WORKS

Improving the customer service process has been a

highly prominent research topic in recent years. This

aspect has aroused deep interest among authors, who

have addressed this challenge in various contexts,

such as the health field, electronic commerce, and the

insurance sector, among others. Therefore, in this

section we present an overview of the different types

of customer service chatbots, as well as the

techniques used in their implementation and the

characteristics that influence the improvement of the

quality of assistance provided to users. The purpose

of this section is to provide a solid understanding of

the current situation and contribute to the

advancement of research in this field.

Firstly, the results obtained by customer service

chatbots are presented. In the study by El-Ansari and

Beni-Hssane (2023), a personalized customer service

chatbot was implemented in e-commerce applications

with sentiment analysis features. The authors

developed this chatbot aiming to enhance the user's

intention recognition capability, considering their

emotions to enhance their experience. In its

development, they employed sentiment analysis

techniques (VADER and BERT) and entity

recognition (NER). In their experiment, 60

participants interacted with two versions of the

chatbot, and the results showed that participants who

interacted with the personalized version (9.13)

reported higher levels of satisfaction compared to

those who interacted with the non-personalized

version (8.41). These findings demonstrate the

effectiveness of the sentiment analysis

personalization process in improving user

satisfaction. Considering this finding, the present

research will employ the GPT-3.5 model for intention

and emotion recognition, aiming to achieve

analogous results in terms of satisfaction, applying

this methodology in the telecommunications domain.

Similarly, in the research conducted by

Kasinathan et al (2021), a customizable multilingual

chatbot for web applications is developed with the

purpose of enabling SMEs to deploy their customer

support chatbot service. Its functionalities include

live chat, ticketing system, and tracking system. In a

survey conducted with 27 participants out of a total of

50 invitations sent to IT departments of companies in

Selangor, Malaysia, 77.8% of the surveyed

companies expressed satisfaction with the use of the

customer service chatbot. Based on these

functionalities, the integration of a ticketing system

with the chatbot of this solution has been considered,

as the knowledge base used to generate responses has

limitations, implying that sometimes a precise

response that solves the user's problem is not

provided. Endowing the chatbot with the

functionality to generate tickets for technical visits

that culminate in problem resolution in case the

chatbot responses are insufficient contributes to

Sentiment Analysis-Based Chatbot System to Enhance Customer Satisfaction in Technical Support Complaints Service for

Telecommunications Companies

making this type of system more efficient. In this

regard, a ticketing system will be used in this research

through integration with the Airtable system, but not

for generating a ticket with an advisor, but rather for

scheduling a technical visit to the

telecommunications user.

On the other hand, techniques used in chatbot

implementation have been found. In the study by

Santos et al (2022), the Evatalk chatbot is developed

using its own Chatbot Management Process (CMP),

which is responsible for managing the chatbot's

contents. Evatalk consists of three components,

namely: EvaTalk Admin, Data Repository, and

Model Trainer. In the latter, natural language

processing (NPL) was employed to identify the user's

intention. The chatbot was validated in customer

technical support at the Virtual School of

Government of Brazil. Its latest user satisfaction test

reached 32.47% at the satisfied level and 50.29% at

the very satisfied level. That being said, this proposal

will employ NPL to identify the user's intention

during the conversation since GPT-3.5 will be used,

which is not only capable of recognizing the intention

but also of discerning emotion in the context of

telecommunications customer technical support.

Additionally, through a random online

experiment in the study by Adam et al. (2021),

empirical research was conducted to examine the

effect of verbal cues incorporating an

anthropomorphic design and the technique known as

'foot in the door' on the degree of compliance with

requests made by the chatbot to users. In this study,

social response theory and the concept of

commitment-consistency were used as theoretical

foundations to guide the research. For the

development of the chatbot, they used IBM Watson

Assistant's cloud service for natural language

processing and dialogue management. The results of

the experiment demonstrated that an

anthropomorphic design, characterized by warm and

empathetic responses, increases the likelihood of

customers complying with the chatbot's requests, as

95% of the total participants complied with the

chatbot's request compared to when these techniques

were not used, which only represented 63%.

Consequently, in this study, we will apply these

characteristics but adapted to the context of complaint

handling in a telecommunications company, focusing

particularly on customer satisfaction survey

compliance.

Similarly, in the study by Rahman and Watanobe

(2023), examines the possibilities and challenges

offered by ChatGPT in the educational context, both

for students and educators in programming teaching.

It is highlighted that ChatGPT presents multiple

exciting advantages in education, including

personalized feedback, increased accessibility, and

interactive conversations. To investigate this, a

survey was conducted with students and teachers

from different academic levels, focusing on the

impact of ChatGPT on programming learning and

teaching. The survey addressed aspects such as

participant identification, their experience in

programming, the support provided by ChatGPT, its

usefulness in problem-solving, as well as the degree

of satisfaction with the system. The results revealed

that most students (86.7%) found the suggestions

provided by ChatGPT helpful in problem-solving. In

this regard, these characteristics of ChatGPT,

including the ability to address issues and facilitate

interactive conversations, will be incorporated into

the research, adapted to the realm of handling

complaints in a telecommunications company.

Finally, concerning best practices in chatbot

development, certain characteristics have been

identified whose incorporation can contribute to

improving the quality of assistance provided to users.

Furthermore, in the research by Shin et al. (2023), the

idea is proposed that the inclusion of humor by

chatbots can play a fundamental role in their

humanization and, consequently, in improving the

customer experience. To support their hypotheses,

they conducted an experiment involving 117 business

university students. These students interacted with a

chatbot in the context of a complaint about an

incorrect bill issued by a fictitious

telecommunications company. Participants engaged

in two different interaction scenarios with the chatbot:

one where the chatbot used humor and another where

it did not. The results of the experiment indicated that

when humor was incorporated, satisfaction with the

service experienced a significant increase compared

to the scenario without humor (M without humor =

4.06 vs. M with humor = 5.05). With this in mind, the

inclusion of humor will be considered as a relevant

feature in the implementation of the solution, albeit

with a specific focus on managing complaints related

to internet and cable services.

Finally, in the comparative satisfaction study

conducted in Mainland China and Hong Kong by Liu

et al. (2023), the importance of customer service

chatbots being competent in timely issue resolution,

showing empathy towards user concerns, and

ensuring data privacy measures to increase usage

intention is emphasized. This study employed a

mixed-method approach integrating the Delone and

McLean's Information System Success Model with

privacy concerns. The results revealed that in

ICSBT 2024 - 21st International Conference on Smart Business Technologies

Mainland China, the model explained 70.4% of the

variation in satisfaction and 51.4% of the variation in

usage intention. In Hong Kong, the model explained

66.9% of the variation in satisfaction and 56% of the

variation in usage intention. These results highlight

the importance of competence, empathy, and security

in user satisfaction and willingness to use customer

service chatbots in these two regions. With that said,

like previous studies, empathy and a competent

approach will be considered as essential features for

the proposed solution, but with a focus on complaint

handling.

3 CONTEXT

3.1 Preliminary Concepts

The purpose of this section is to explain the necessary

elements for the construction of a customer service

chatbot, such as chatbot technology, natural language

processing, ChatGPT, and sentiment analysis. In this

section, the key concepts used in this work are

presented.

3.1.1 Chatbot

It’s a conversational agent based on artificial

intelligence and machine learning technology that

offers a variety of services through communication

channels, such as text messages. It is designed to

interact with human users in an automated manner

(Jaspin et al., 2023).

3.1.2 Natural Language Processing

It is a technology that focuses on the interaction

between the computer and human language.

Additionally, it is used to analyse large amounts of

information from various sources, such as patent

databases, social networks, or crowdsourcing

platforms, with the purpose of facilitating the search

for promising solutions (Just, 2023).

3.1.3 ChatGPT

It is a pre-trained language model based on artificial

intelligence that utilizes the GPT-3.5 architecture,

developed by the company OpenAI. This model

boasts remarkable capabilities in the field of natural

language processing, enabling it to understand and

respond to human language in real-time (Kocon et al.,

2023

3.1.4 Sentiment Analysis

Sentiment analysis (SA) is a function of natural

language processing aimed at extracting sentiments

and evaluations expressed in texts. Its essence lies in

the ability to discern how sentiment is manifested in

a text (Bernabé et al., 2020).

3.2 Method

In this section, a detailed analysis of the design of the

chatbot based on Natural Language Processing

technology GPT 3.5 is presented, which enables the

application of sentiment analysis, humor, warmth in

conversations, and empathy. The Figure 1 illustrates

the 4-layer architecture of ChatTelecom. Throughout

the description of each layer, the five fundamental

components that constitute the chatbot system will be

addressed, each of which assumes a singular and

essential function in the operation of the solution.

3.2.1 User Layer

This layer serves as the medium through which users

interact with the chatbot. On one hand,

telecommunication clients can use the WhatsApp

application via a cell phone or laptop. On the other

hand, technical support personnel handle complaints

through a PC.

3.2.2 Presentation Layer

The main component within this layer is

Whatsapp/front end. This component is a messaging

platform developed by WhatsApp that allows

companies to interact with their customers through

this tool. WhatsApp was chosen as the front end

because it is one of the main customer service

channels in Peru across various industries. The

presentation layer serves as an essential link between

ChatTelecom, and conversations held via WhatsApp.

On one hand, telecommunication service clients

submit their complaints through WhatsApp

messages. Lastly, when a complaint requires

intervention from an advisor, i.e., a member of the

technical support team, they handle and manage the

complaint using the specialized version of WhatsApp

Business. It’s important to clarify that the role of

technical support staff is to act as an intermediary in

the communication, intervening only when the

chatbot is unable to resolve the telecommunications

customer's issue.

Sentiment Analysis-Based Chatbot System to Enhance Customer Satisfaction in Technical Support Complaints Service for

Telecommunications Companies

3.2.3 Service Layer

In the chatbot service process, it requests personal

information from the internet and/or cable television

service holder. Then, data validation proceeds with

diagnosing the reported issue, classifying the

breakdown. Once the type of breakdown the customer

is facing is identified, multimedia content (images

and PDFs) is requested to choose the most

appropriate resolution step. Furthermore, to assist the

customer in resolving their issue based on the

chatbot's instructions, multimedia content will be

sent. Hence, a file server is available to store these

documents. In order to ensure better customer service,

the GPT 3.5_Sentiment Analysis component allows

adapting the conversation with the customer based on

the detected emotion and applying anthropomorphic

characteristics (empathy, warmth in conversations,

and humor). Below is the application of its

subcomponents.

Component Sentiment Analysis

The sentiment analysis component provided by the

GPT 3.5 model was utilized. Initially, the choice of

GPT 3.5 technology was made because it allows

recognizing the user's intention during the

conversation with an accuracy of 85.5%, possesses a

massive corpus of data, and has the capability to adapt

to different scenarios (Panda & Kaur, 2023; Zhu et

al., 2023). However, since the problem addressed

involves user emotional states being relevant to the

service, it was necessary to fully leverage this model

to enhance the user experience. The sentiment

analysis component of this model was used as it

allows adapting the chatbot's responses based on the

recognized emotions, which can be expressed through

texts or emojis. The application of sentiment analysis

in the solution was carried out through the

customization of prompts, where it was specified to

recognize the user's emotion in order to adapt its

response based on the detected emotion.

Application of Humor, Empathy, and

Warmth in Conversations

The implementation of the chatbot considered these 3

anthropomorphic characteristics. Empathy was

prioritized to offer users a closer and more

personalized service experience. Additionally,

warmth in responses was incorporated, recognizing

that friendly responses foster more effective

communication by making customers feel

comfortable expressing their concerns. Furthermore,

the use of humor was explored, acknowledging that

well-employed humor can transform customer

service interaction into a more enjoyable and

memorable experience. These characteristics were

defined as sections within the prompts of GPT 3.5.

For their implementation, specific prompts within

GPT 3.5 were used to guide the model to reflect

empathy, maintain warmth in conversations, and

employ relevant humor according to the detected

emotion of the user. These components were refined

during the chatbot training, which included

keywords, key phrases, machine learning

functionality provided by the platform, and constant

interaction with users.

Figure 1: Architecture of the proposed ChatTelecom.

ICSBT 2024 - 21st International Conference on Smart Business Technologies

Finally, if the chatbot is unable to resolve the

issue due to its complexity, the case is redirected to

an advisor. This process involves the technical

support staff continuing the resolution of the reported

problem. During this process, the chatbot provides

tools that facilitate the advisor's work, such as a

summary of the interaction between the user and the

chatbot, as well as the representation of the user's

detected emotions through emojis.

3.2.4 Data Layer

In the data layer, data extraction is performed from

the system Airtable. This data includes elements such

as names, surnames, the National Identity Document

(DNI) number, and the IP address corresponding to

the internet and/or cable service holder. This

extraction process is carried out with the primary

purpose of validating the user's identity at the

beginning of the assistance process. On the other

hand, the client's IP address is used to simplify and

expedite the assistance procedure by the advisor.

Thanks to the combination of these components

within the ChatTelecom solution based on GPT 3.5,

this research aims to achieve a 3% increase in

customer satisfaction during the process of

addressing complaints related to internet and cable

television services within the telecommunications

companies.

4 EXPERIMENTS

In this section, we will discuss the experiments

carried out in the project, addressing both their design

and execution, and present the conclusions derived

from the results obtained throughout this process.

4.1 Experimental Protocol

The chatbot was developed using Voiceflow as this

platform facilitated integration with GPT-3.5 and

WhatsApp. On the other hand, for the experiments

with ChatTelecom, specific equipment was required

such as a PC with at least 8 GB of RAM and a 2.5

GHz processor, or a mobile device with at least 4 GB

of RAM for a smooth experience.

The 50 participants, aged between 20 and 60 years

old, reside in the northern region of the capital of Peru

and are subscribers to internet and/or cable television

services provided by a telecommunications company.

User satisfaction was evaluated using a Likert scale

from 1 to 5, and the chatbot's ability to recognize

emotions was considered across 3 different scenarios.

35 participants were randomly selected for scenarios

1 and 3, with 20 assigned to scenario 1 and 15 to

scenario 2. On the other hand, the remaining 15 were

assigned to scenario 2, as they only had internet

service. Below is the formula for the American

Customer Satisfaction Index (ACSI), which was used

to evaluate the user satisfaction index:

ACSI = 

∗ S



−D





+D

(1)

Where:

 N is the total number of responses in the

survey.

 n is the number of responses considered in the

calculation.

 𝑆



represents the ratings of each user.

 D is the benchmark satisfaction rating for the

industry.

The experiment begins with the user presenting a

complaint about internet and/or cable television

service to the customer technical support chatbot. It

starts by welcoming the user and proceeds to request

necessary data for resolving the complaint. In the first

scenario, the user informs the chatbot that their issue

is the lack of clarity in the cable television channels.

Simultaneously, the chatbot identifies that the user

entered with a feeling of anger. Subsequently, after a

series of specific questions related to the complaint

process, the chatbot successfully resolves the

incident, concluding with the delivery of a

satisfaction survey regarding the assistance provided.

In the second scenario, the user informs the chatbot

that their problem is the loss of internet service.

Simultaneously, the chatbot identifies that the user

entered with a feeling of impatience as they requested

a prompt solution. During the diagnosis of the fault,

the chatbot asks the customer questions to identify the

cause of the problem. Finally, the chatbot concludes

that there is a router equipment misconfiguration

issue, so it informs the customer that it is not able to

resolve this type of faults but could refer them to an

advisor who could address their case. Lastly, in the

third scenario, the user details to the chatbot that their

problem is the loss of cable television service.

Simultaneously, the chatbot identifies that the user

entered feeling concerned about their problem. After

a series of questions aimed at resolving the problem,

the chatbot determines that it is necessary to schedule

a technical visit. Therefore, the problem is referred to

a specialized advisor. The advisor concludes the

assistance and proceeds to issue a ticket for the

technical visit. Upon concluding the conversation

flow with the chatbot, it generates a customer

satisfaction survey that includes the following

Sentiment Analysis-Based Chatbot System to Enhance Customer Satisfaction in Technical Support Complaints Service for

Telecommunications Companies

questions:

 How satisfied are you with the assistance you

received today? 1: Very dissatisfied, 2:

Dissatisfied, 3: Neutral, 4: Satisfied, 5: Very

satisfied.

 How would you rate our ability to understand

how you feel during the conversation? 1: Very

low ability, 2: Low ability, 3: Average ability,

4: High ability, 5: Very high ability.

It’s important to highlight that, according to the

analysis of the studies conducted during the research,

no studies have been found that offer a comparison

with the emotional recognition capability indicators

and the American Customer Satisfaction Index

(ACSI) used in this research. Therefore, it is proposed

to consider these indicators as new.

4.2 Results

In this section, the results of the satisfaction metrics

and emotional recognition capacity of the user

obtained from the experiments are presented.

On one hand, the experiment results revealed that

the average rating given to the first satisfaction

question is 4.4 on a scale ranging from 1 to 5.

Additionally, the ACSI yielded a score of 77.9,

surpassing the threshold of 75 points established as

the quality standard in customer service in Peru

(OSIPTEL, 2021).

𝐴

𝐶𝑆𝐼 = 

∗

222 − 75

 + 75

(2)

ACSI = 77.9

Figure 2: User Satisfaction Rate.

On the other hand, the user satisfaction rate

achieved was 86% compared to other customer

service chatbots mentioned in the related works

section, as shown in Figure 2.

The results from these three scenarios were

grouped to measure user satisfaction because,

although the resolution methods differ, the goal in all

cases is the same: to resolve the customer’s problem

and assess their satisfaction with the entire support

process. Therefore, by grouping the scenarios, an

integral view of the system's overall effectiveness is

obtained, allowing for the identification of areas for

improvement for both the chatbot and human

intervention. Additionally, this grouping reflects a

more realistic user experience, where satisfaction

depends not only on who resolves the problem, but on

how the overall support service is perceived.

On the other hand, based on the second question

to evaluate the chatbot's ability to recognize how the

user feels during the conversation, positive results

were obtained.

Figure 3: User emotional recognition rate.

According to Figure 3, most participants stated

that their emotional state is taken into consideration

during interactions with the chatbot. 56% of

respondents rated this capability as very high, while

36% rated it as high. On the other hand, 8% rated it

as having an average capacity to understand their

emotional state. It's important to highlight that the

user perceives that the chatbot recognizes their

emotions due to its ability to respond appropriately

and sensitively to their emotional expressions during

the interaction. This is reflected in the chatbot's

responses, the use of language, and the relevance of

the solutions offered in relation to the emotions

expressed by the user. These results are largely

attributed to the sentiment analysis component of the

system, which recognizes emotion and reflects it

through emojis to mitigate it. Additionally, the

application of empathy, warmth in conversations, and

humor is considered to have a significant influence.

5 CONCLUSION

The application of humor, warmth in conversations,

empathy, and sentiment analysis within the chatbot

solution represents an advancement in the field of

customer service. On one hand, it is evident that the

sentiment analysis component of the GPT-3.5 model

ICSBT 2024 - 21st International Conference on Smart Business Technologies

has been crucial in emotional recognition, as 56% of

respondents rated the chatbot with a very high

capacity to understand their emotional state. On the

other hand, the combination of anthropomorphic

features and sentiment analysis proved favourable for

improving satisfaction, achieving a satisfaction rate

of 86%. In conclusion, these components emerge as

crucial factors for enhancing quality in the complaint

handling process in the technical support domain for

telecommunications companies. This understanding

is supported by the fact that the satisfaction index

obtained is 77.9, thus surpassing the minimum

threshold of 75 established by the

telecommunications industry in Peru as a parameter

to ensure quality in the customer service process.

While the results obtained were indeed positive,

we believe that there is room for additional

improvements. We suggest integrating the chatbot

with the customer network management systems of

the telecommunications company, as well as utilizing

an additional specific sentiment analysis component

alongside the response model. Additionally, it is

recommended to periodically update the chatbot's

knowledge base, as the GPT-3.5 component handles

data that may be outdated. This update should include

the incorporation of new cases reported by customers,

recent feedback, and current problems to provide

more accurate answers to current problems.

Furthermore, we estimate that a broader survey

sample would have contributed to a more

representative evaluation of the chatbot's

effectiveness.

With the continuous progress of technology, it’s

expected that chatbots will acquire greater

intelligence and capability to address a wider range of

tasks. This advancement, consequently, would

positively impact customer satisfaction and overall

improvement of a company's service process.

Therefore, we consider it imperative to continue

conducting research employing current NLP models.

Additionally, we express interest in future research

evaluating similar components to those used in this

work but applied in other customer service contexts,

where user emotions also influence the service.

FINANCING

Peruvian University of Applied Sciences / UPC-

EXPOST-2023- 2.

ACKNOWLEDGEMENTS

To the Research Department of the Peruvian

University of Applied Sciences for the support

provided for the completion of this research work

through the UPC-EXPOST-2023-2 incentive.

REFERENCES

Adam, M., Wessel, M., & Benlian, A. (2021). AI-based

chatbots in customer service and their effects on user

compliance. Electronic Markets, 31(2), 427–445.

https://doi.org/10.1007/s12525-020-00414-7

Akhtar, M., Neidhardt, J., Werthner, H., (2019). The

Potential of Chatbots: Analysis of Chatbot

Conversations. IEEE Xplore. https://doi.org/10.1

109/CBI.2019.00052

Behera, R., Bala, P., & Ray, A. (2021). Cognitive Chatbot

for Personalised Contextual Customer Service: Behind

the Scene and beyond the Hype. Information Systems

Frontiers, 1–21. https://doi.org/10.1007/S10796-021-

10168-Y

Bernabé, J., Tejeda, A., Herce, J., Porcel, C., & Herrera, E.

A context-aware embeddings supported method to

extract a fuzzy sentiment polarity dictionary. (2020).

Knowledge-Based Systems, 19029. https://doi.org/

10.1016/j.knosys.2019.105236

Ehrenpreis, M., & DeLooper, J. (2022). Implementing a

Chatbot on a Library Website. Journal of Web

Librarianship, 16(2), 120–142. https://doi.org/10.1080

/19322909.2022.2060893

El-Ansari, A., & Beni-Hssane, A. (2023). Sentiment

Analysis for Personalized Chatbots in E-Commerce

Applications. Wireless Personal Communications,

129(3), 1623–1644. https://doi.org/10.1007/s11277-

023-10199-5

Jaspin, K., Sanjeev, S., & Shyam, S. (2023). Chatbot for

Trading Cryptocurrency. Proceedings of the 2nd IEEE

International Conference on Advances in Computing,

Communication and Applied Informatics.

https://doi.org/10.1109/ACCAI58221.2023.10200405

Just, J. (2023). Natural language processing for innovation

search – Reviewing an emerging non-human

innovation intermediary. Technovation, 129.

https://doi.org/10.1016/j.technovation.2023.102883

Kasinathan, V., Mustapha, A., & Bin, C. (2021). A

customizable multilingual chatbot system for customer

support. Annals of Emerging Technologies in

Computing, 5(5), 51–59. https://doi.org/10.33

166/AETiC.2021.05.006

Kocon, J., Cicheki, I., Kaszyca, O., Kochanek, M., Szydlo,

D., Baran, J., Bielaniewicz, J., Gruza, M., Janz, A.,

Kanclerz, K., & Kocon, A., Koptyra, B. (2023).

ChatGPT: Jack of all trades, master of none.

Information Fusion, 99. https://doi.org/10.1016/

j.inffus.2023.101861

Sentiment Analysis-Based Chatbot System to Enhance Customer Satisfaction in Technical Support Complaints Service for

Telecommunications Companies

Liu, Y., Hu, B., Yan, W., & Lin, Z. (2023). Can chatbots

satisfy me? A mixed-method comparative study of

satisfaction with task-oriented chatbots in mainland

China and Hong Kong. Computers in Human Behavior,

143. https://doi.org/10.1016/J.CHB.2023.107716

Lu, Q., Sun, X., Long, Y., Gao, Z., Feng, J., & Sun, T.

(2023). Sentiment Analysis: Comprehensive Reviews,

Recent Advances, and Open Challenges. IEEE

Transactions on Neural Networks and Learning

Systems, 1-121. https://doi.org/10.1109/TNNLS.2023.

3294810

Ngai, E., Lee, M., Luo, M., Chan, P., & Liang, T. (2021).

An intelligent knowledge-based chatbot for customer

service. Electronic Commerce Research and

Applications, 50. https://doi.org/10.1016/j.elerap.

2021.101098

Organismo Supervisor de Inversión Privada en

Telecomunicaciones (2022). Reclamos por averías

[Conjunto de datos]. Repositorio de Datos Abiertos del

OSIPTEL. https://acortar.link/MxzMqL

Organismo Supervisor de Inversión Privada en

Telecomunicaciones. (2021). Ranking de la calidad de

atención al usuario. https://bit.ly/40yxj06

Panda, S., & Kaur, N. (2023). Exploring the viability of

ChatGPT as an alternative to traditional chatbot

systems in library and information centers. Library Hi

Tech News, 40(3), 22–25. https://doi.org/10.110

8/LHTN-02-2023-0032/FULL/XML

Santos, G., De Andrade, G., Silva, G., Duarte, F., Costa, J.,

& DeSousa, R. (2022). A Conversation-Driven

Approach for Chatbot Management. IEEE Access, 10,

8474–8486. https://doi.org/10.1109/ACCESS.2022.

3143323

Shin, H., Bunosso, I., & Levine, L. (2023). The influence

of chatbot humour on consumer evaluations of services.

International Journal of Consumer Studies, 47(2), 545–

562. https://doi.org/10.1111/ijcs.12849

Spicework. (2018, 2 de abril). Data snapshot: AI Chatbots

and Intelligent Assistants in the Workplace.

Recuperado el 26 de agosto de 2023, de

https://community.spiceworks.com/blog/2964-data-

snapshot-ai-chatbots-and-intelligent-assistants-in-the-

workplace

Zhu, S., Ma, T., Rong, H., & Al-Nabhan, N. (2023). A

Personalized Multi-Turn Generation-Based Chatbot

with Various-Persona-Distribution Data. Applied

Sciences, 13(5). https://doi.org/10.3390/APP13053122.

ICSBT 2024 - 21st International Conference on Smart Business Technologies