Requirements Engineering Challenges and Techniques in Building

Chatbots

Geovana R. S. Silva

and Edna Dias Canedo

Department of Computer Science, University of Bras

ılia (UnB), P.O. Box 4466, Bras

ılia–DF, Zip code 70910-900, Brazil

Keywords:

Chatbot, Conversational Agents, Software Engineering, Requirements Engineering, Requirements Elicitation.

Abstract:

The proper execution of requirements engineering processes can be decisive for the success of software devel-

opment and, for chatbots, it is no different. Chatbots have been gaining a lot of space, especially in customer

service. Requirements engineering processes for chatbots are as hard to perform as for any other machine-

learning system and the conversational nature makes it even harder. Taking this into consideration, in this

work we survey chatbot practitioners to unveil the requirements elicitation and documentation techniques they

have been using in the industry, besides the challenges they face while going through this process. Responses

show that elicitation techniques are not much far from techniques used in other ﬁelds, but for documentation

techniques that are new forms of documentation such as conversation ﬂows. Moreover, meeting stakeholder’s

requirements and managing information exchange are their biggest challenges in eliciting and documenting

chatbot requirements.

1 INTRODUCTION

AI-based systems are crossing the academic barrier to

be increasingly used due to two major factors: the in-

creasing availability of big data and hardware acceler-

ators (Soni et al., 2019). The automation capability of

AI services is being especially used in customer ser-

vices to streamline user service (Khan and Das, 2018)

and are increasingly being better accepted by users

(Xu et al., 2020). Chatbots are one example of how

AI systems are being used in customer services.

Chatbots are intelligent agents powered by ma-

chine learning algorithms to mimic human behav-

ior, and users tend to recur to them because of the

ease, speed, and convenience of chatting with a chat-

bot (Brandtzaeg and Følstad, 2017). Since technol-

ogy geared towards conversational agents is becom-

ing more trustworthy for commercial use, companies

have been using chatbots as part of their customer ser-

vice workforce. Companies recur to this kind of tech-

nology for the following main reasons(Zhang et al.,

2021): improve customer service experience, reduce

cost and resource requirements, and drive digitaliza-

tion.

Requirements engineering processes for real-

https://orcid.org/0000-0002-0304-0804

https://orcid.org/0000-0002-2159-339X

world AI-based systems require different approaches

than traditional engineering because of their data-

driven nature (Kostova et al., 2020). Speciﬁcally for

machine-learning (ML) systems, requirements engi-

neering processes are challenged by the lack of ex-

plainability of ML systems, difﬁculties in insuring

freedom of discrimination, legal constraints over data,

and eliciting data requirements (Vogelsang and Borg,

2019).

These challenges are also faced by chatbot prac-

titioners, in addition to the difﬁculties of meeting

requirements related to the conversational nature of

chatbots. Although chatbots share the responsibility

to engage in a conversation, they can become very dif-

ferent from one another to meet the needs of the brand

they represent (Shevat, 2017). This reﬂects on re-

quirements, which must be of good quality to achieve

the business goals with the chatbot. Traditionally,

software development teams go from requirements

to features. In the context of conversational agents,

teams go from requirements to meaningful dialogues.

The most valuable delivery for chatbot applications is

conversational ﬂows that solve a user’s doubt or prob-

lem. In fact, if requirements focus ﬁrstly on conversa-

tion experience, technological architecture will be de-

veloped to ﬁt the conversation ﬂow and not the other

way around.

In this paper, we have conducted an online-based

180

Silva, G. and Canedo, E.

Requirements Engineering Challenges and Techniques in Building Chatbots.

DOI: 10.5220/0010801800003116

In Proceedings of the 14th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2022) - Volume 1, pages 180-187

ISBN: 978-989-758-547-0; ISSN: 2184-433X

survey with chatbot practitioners to discover the main

tools and techniques used in requirements engineer-

ing processes in building chatbots. We are also in-

terested in getting to know how practitioners perceive

requirements processes in this context and what chal-

lenges they face.

2 BACKGROUND

The success of software highly depends on meet-

ing the imposed requirements of stakeholders, which

also depends on a good deﬁnition of requirements.

For that to happen, software engineering comprises

requirements engineering. According to (Pressman,

2009), requirements engineering is the spectrum of

tasks and techniques that lead to a clear understanding

of the requirements, building a bridge to design and

construction. There are four main activities in the re-

quirements engineering process (Sommerville, 2010):

1. Feasibility study: investigate if the proposed sys-

tem is going to be cost-effective and the result deter-

mines if the process will continue to the next steps;

2. Elicitation and analysis: work with stakeholders to

gather information and use a set of techniques to dis-

cover requirements; 3. Speciﬁcation: the process of

writing down requirements in a structured document;

and 4. Validation: checks the requirements for real-

ism, consistency, and completeness.

In this work, we focus on elicitation and speci-

ﬁcation. Elicitation activities are tied to clients and

customers because they hold valuable information for

discovering requirements. Speciﬁcation activities are

tied to lots of documents that formalize and model de-

cisions of previous activities. We are especially inter-

ested in the techniques used by chatbot practitioners

when executing these activities besides the challenges

they face when doing it.

2.1 Requirements for Chatbots

Chatbot practitioners face additional challenges in

chatbot requirements engineering processes since

chatbots “have speciﬁc UX requirements related to

the way the information is presented, the mean of in-

teraction, whether its text, buttons or speech” (Fadhil,

2018). Another challenge of chatbots is designing for

open-ended conversations (Brandtzaeg and Følstad,

2018) because conversational requirements must be

broad enough to cover all conversation possibilities.

Therefore, for this context, there are two new classes

of requirements: conversation and data, which are not

seen in traditional requirements engineering. Deﬁn-

ing data requirements includes deﬁning what the AI

model should detect, such as age groups and language

(Akkiraju et al., 2020), and, if relying on open-source

data, the ﬁltering criteria for selecting high-quality

data (Vogelsang and Borg, 2019). Conversational re-

quirements are all about deﬁning how it communi-

cates with users, including its persona, formality of

language, visual aids, and conversation shortcuts.

Having well-deﬁned conversational requirements

is important because the customer’s quality percep-

tions are increased if the brand’s language is simi-

lar to the customer’s expectation (Jakic et al., 2017).

Therefore, in the development of chatbots, conversa-

tions must be seen as the object of design (Følstad

and Brandtzæg, 2017). Moreover, diving into know-

ing your target audience will dictate how your chat-

bot communicates, since ”younger users may be par-

ticularly sensitive to playful and emotionally engag-

ing chatbots, whereas older users may be preoccu-

pied with the efﬁciency and effectiveness of chatbots”

(Følstad and Brandtzaeg, 2020).

Another important aspect to consider in the elici-

tation process is if there are existing sources of data

with interactions between the service provider and

potential chatbot users, which can be phone calls,

emails, chat, social media, and customer self-service

(Singh et al., 2019). These sources already contain

some requirements for conversation content, espe-

cially if the chatbot is being developed to replace one

or more of these traditional channels of interaction.

Missing these speciﬁc classes of requirements

hurt the chatbot’s inclusiveness capacity and cause

unimaginable damage to brand images. One exam-

ple is Microsoft’s Tay Bot released in 2016, which

started to post offensive tweets when learning from

user’s tweets (Neff and Nagy, 2016). In a research

led by The Washington Post, it was revealed that

Amazon’s and Google’s smart speakers work best for

white, highly educated, upper-middle-class Ameri-

cans. (Harwell, 2018). These failures are not uncom-

mon and because of that conversation design has been

highlighted by big companies as an important part of

chatbot development. Google released its own guide

for conversation design (Google, 2021), as well as

IBM (IBM, 2021) and Amazon (Amazon, 2021). All

of these guides put the user at the center of develop-

ment. Google’s guide has a speciﬁc section for gath-

ering requirements that is entirely focused on thinking

as the end-users and deﬁning who are they, what their

needs are, their context, and their journeys through

conversations.

Requirements Engineering Challenges and Techniques in Building Chatbots

181

2.2 Related Works

To our knowledge, there are no works addressing

speciﬁcally requirements engineering processes in

building chatbots. However, there are works ap-

proaching challenges and techniques in a broader

scope that includes chatbots, such as artiﬁcial intelli-

gence and machine learning. In (Vogelsang and Borg,

2019), four data scientists are interviewed to under-

stand how machine learning experts approach elicita-

tion, speciﬁcation, and validation of requirements and

expectations. The results show that speciﬁc character-

istics of machine learning trigger changes in require-

ments engineering processes, such as new kinds of re-

quirements and the tailoring of traditional approaches.

The work of (Habibullah and Horkoff, 2021) fo-

cus on non-functional requirements of machine learn-

ing systems. They interview ten engineer practition-

ers to understand what are the challenges of dealing

with non-functional requirements for machine learn-

ing in the industry. In the light of their results, they

highlight how difﬁcult it is to gather quality attributes

for machine learning systems.

(Mendonc¸a et al., 2021) conducted a literature re-

view of requirements engineering processes of agent-

oriented software engineering. They focus on dis-

covering how requirements engineering was adapted

for the development of multi-agent systems. (Be-

lani et al., 2019) identiﬁed the requirements engi-

neering challenges when applied to the development

of AI-based complex systems. After gathering chal-

lenges presented in related works in the ﬁelds of arti-

ﬁcial intelligence and software engineering, they pro-

pose the RE4AI taxonomy that outlines the challenges

posed to requirements engineering towards building

AI-based complex systems.

In (Wilhjelm and Younis, 2020), a systematic

process is developed for eliciting security require-

ments of machine learning-based systems. (Ishikawa

and Matsuno, 2020) discusses the impact of machine

learning uncertainty on existing requirements engi-

neering methods such as goal-oriented requirements

elicitation.

3 RESEARCH STRATEGY

In this work, we have investigated how requirements

engineering processes are held in the industry of chat-

bots by surveying Brazilian practitioners. This survey

intends to answer the following research questions:

RQ.1 What are the techniques and tools used by prac-

titioners to elicit and document chatbot require-

ments?

RQ.2 Are the existing tools and techniques of re-

quirements engineering suitable for chatbot de-

velopment?

RQ.3 How important do practitioners perceive re-

quirements engineering processes in the con-

text of chatbots?

RQ.4 What challenges do practitioners encounter

when eliciting and documenting chatbot re-

quirements?

In summary, the research questions are focused on

discovering practices and challenges for eliciting and

document chatbot requirements. We hypothesize that

results will differ from requirements engineering pro-

cesses from software in general because of the con-

versational nature of chatbots, which may require dif-

ferent approaches. For answering research questions,

we made a survey of 12 questions to share with chat-

bot practitioners which are transcribed and have their

results collected in Section 4.

The survey was online-based and the data collec-

tion followed two approaches. The ﬁrst one was to

share the survey link through professional networks

and rely on person-to-person disclosure. Since we

were looking for practitioners of a niche, we did not

expect to collect many responses from this ﬁrst ap-

proach. After this ﬁrst attempt, we proceeded to ﬁnd

these practitioners through professional proﬁles that

revealed their chatbot background and invited them to

answer our survey. We tried to invite Brazilian practi-

tioners from different companies and experiences, to

gather broader responses.

4 SURVEY RESPONSES

The survey received a total of 22 responses from chat-

bot practitioners of different backgrounds. In the next

paragraphs, we show results from each question.

Q1. What is the nature of your company? Re-

spondents could choose between ”Public” or ”Pri-

vate”. 17 (77.3%) were from private companies

whereas only 5 (22.7%) were from the public sec-

tor. The small number of practitioners from the public

sector seems to be circumstantial since chatbots have

been widely used in the public sector (Petriv et al.,

2019; Adnan et al., 2020; Sun and Medaglia, 2019).

Q2. Which role have you assumed in chatbot

projects? Respondents had a set of chatbot-related

roles to choose from or the ”other” option for the ones

that did ﬁt one of them. Developers represented most

of the respondents (63.6%), followed by tech leaders

(18.2%) and conversation designers (13.6%). One re-

spondent marked the ”other” option and justiﬁed in

ICAART 2022 - 14th International Conference on Agents and Artiﬁcial Intelligence

182

the open ﬁeld of Q12 that he assumed many of the

listed roles for similar periods.

Q3. What software development methodology

does your chatbot development team use? The ma-

jority of respondents work with the Scrum method-

ology (68.2%), followed by Kanban (18.2%) and Ag-

ile(9.1%). One respondent marked ”We have not used

a structured methodology”.

Q4. What requirements elicitation techniques

have your team used to develop chatbots? Figure

1 shows all the techniques marked by practitioners.

There were predeﬁned options for them to mark but

they could add new options if there were techniques

that they have used but were not listed. Brainstorming

is by far the most used technique by chatbot practi-

tioners (68.2%).

Figure 1: Elicitation techniques used by chatbot practition-

ers.

Q5. What requirements documentation tech-

niques have your team used to develop chatbots?

Figure 2 shows all the techniques marked by practi-

tioners. In accordance with the previous questions,

respondents could choose from predeﬁned options or

list new ones. The interesting result is that chatbot re-

quirements documentation is mostly done with con-

versation ﬂows (81.8%), which is a particular option

for the ﬁeld we are surveying in contrast with other

software ﬁelds. In contrast, user stories are widely

used in diverse software ﬁelds and are also used by

59.1% of chatbot practitioners. Moreover, user stories

are not unfamiliar for agent-driven software architec-

tures as seen in (Wautelet et al., 2016).

Q6. Which tools have your team used to elicit

or document chatbot requirements? Since there are

Figure 2: Documentation techniques used by chatbot prac-

titioners.

Table 1: Tools that chatbot practitioners have used in elicit-

ing and documenting chatbot requirements.

Tool Description Citations

Bizagi Tool for modeling business process

maps with structured language.

Draw.io Online diagram software with support

for various notations and free model-

ing.

Figma Online UI tool to create, collaborate,

prototype, and handoff.

Google

Forms

Helps in the creation of questionnaires

to do surveys and collect data.

LucidChart Online diagram software with support

for various notations and free model-

ing.

Microsoft

Word

Program for creating natural language

documents, commonly known as word

processor.

Microsoft

Excel

Program for organizing and formatting

spreadsheets with mathematical formu-

las and computerization.

Miro Digital whiteboard for distributed

teams collaboration.

Whimsical Visual workspace for combining

ﬂowcharts, wireframes, sticky notes,

mind maps and documents.

XMind Tool dedicated exclusively to creating

mind maps.

countless tools for supporting the requirements elici-

tation and documentation, we opted for an open and

optional ﬁeld. We gathered 16 responses that exposed

multiple tools used in this process. Table 1 shows

tools cited by respondents. They are all very gen-

eralist tools that can support many of the techniques

listed, even for informal and intermediary documen-

tation.

Q7. In the requirements listing of your chatbot

project, were there non-functional requirements?

Although 68.24% of practitioners responded ”Yes”, it

is particularly alarming that more than 30% did not

have non-functional requirements considering chat-

Requirements Engineering Challenges and Techniques in Building Chatbots

183

bots require speciﬁc usability and user-experience

standards to carry on meaningful conversations.

Q8. Do you think it is necessary to have speciﬁc

approaches for eliciting and documenting chatbot

requirements? Previous questions received scattered

responses, especially Q6. This question’s results re-

inforce the need for adaptation in eliciting and doc-

umenting chatbot requirements since 77.3% marked

”In part, because existing techniques, processes and

tools need to be adapted to the context of chatbots”.

13.6% responded ”Yes” and 9.1% responded ”No”.

Q9. Do you think that requirements elicitation

and documentation processes for chatbots are ne-

glected? Responses were well distributed among op-

tions as shown by Figure 3, probably because their

companies treat requirements differently. However,

responses are more concentrated on the negative spec-

trum.

Figure 3: Do you think that requirements elicitation and

documentation processes for chatbots are neglected?

Q10. Do you think the non-functional require-

ments for chatbots are neglected? In comparison

with Q9, practitioners think that non-functional re-

quirements are more neglected than the processes

in general since responses were: ”Very neglected”

(36.4%), ”Moderately neglected” (50.0%), ”Slightly

neglected” (13.6%), and ”Not neglected” (0%).

Q11. How important is it for you to have a

well-structured requirements engineering process

for chatbot development? Figure 4 shows the dis-

tribution of responses. Although the previous ques-

tions seems to indicate that chatbot requirements are

not being treated as they should, 77.3% of the re-

spondents see well-structured requirements engineer-

ing processes as ”Very important”. The one respon-

dent that marked ”Not important” is a conversation

designer. In (Franch et al., 2020), designers are one of

the roles who are less favorable towards requirements

engineering research, which can corroborate the re-

sponse from this respondent, although the other two

designers considered it ”Very important”.

Q12. Which challenges have you or your team

faced in eliciting or documenting the requirements

of a chatbot? This question was optional, open, and

received 12 responses. Since open responses leave

Figure 4: How important is it for you to have a well-

structured requirements engineering process for chatbot de-

velopment?

space for embracing a lot of subjects, we analyzed

responses and classiﬁed the challenges practitioners

cited directly or indirectly in their responses. Table 2

shows this classiﬁcation.

5 DISCUSSION

When grouping responses by the organization’s na-

ture or the practitioner’s role, we did not observe any

particular phenomenon and responses were very di-

verse within these groups. This may indicate that

techniques and processes are probably more linked to

the speciﬁc processes of each company, either public

or private, than roles or the organization’s nature.

Regarding the elicitation techniques used by prac-

titioners, results are not very different from the tech-

niques used in other ﬁelds of requirements engineer-

ing. Brainstorming, interview, and prototyping ap-

peared as at the top 5 of most used techniques in the

context of startups (Raﬁq et al., 2017), mobile ap-

plications (Dar et al., 2018), IT ukranian profession-

als (Gobov and Huchenko, 2020), Internet of Things

(Lim et al., 2018) and software development in gen-

eral (Alﬂen and Prado, 2021).

However, we found some differences in our ﬁnd-

ings compared with the top techniques in other ﬁelds.

First, some of our documentation techniques ap-

peared as elicitation techniques, which were user sto-

ries and use cases. In works in which this happened,

they appeared in the top 5 techniques as well. More-

over, interviews lead most used techniques in all of

the 5 works, on the other hand, brainstorming is the

most popular technique among chatbot practitioners.

In relation to documentation techniques, user

stories are the most used documentation technique

in Agile projects (Jarzbowicz and Połocka, 2017;

Jarzbowicz and Sitko, 2019). Similarly, as most of

our respondents work with Scrum, an Agile method-

ology, user stories appeared as the second most used

technique among them. What stands out in this part

of our results is that conversation ﬂows are the most

ICAART 2022 - 14th International Conference on Agents and Artiﬁcial Intelligence

184

Table 2: Challenges in eliciting or documenting chatbot re-

quirements that appeared in practitioners’ responses.

Challenge Transcribed excerpts from re-

sponses

Difﬁculties in meet-

ing stakeholders’

needs

”Understanding what the

client/user wants [...]”

”Make a chatbot for the end-user

which serves the end-user [...]”

”The transfer of information be-

tween the client and the end user is

a big challenge [...]”

”Problem prioritization, in terms of

aligning stakeholders’ expectations

with actual stakeholders’ needs

[...]”

”Clients’ objectivity with what they

want from the bot [...]”

”In general, it’s the client’s under-

standing. It takes me 1 month to

make a bot, and the client takes 2

years to get what he really wanted

[...]”

”Sincerity of users [when testing]

[...]”

Poorly documented

requirements

”Always rework for requirements

that were not well deﬁned [...]”

”Often when the area responsible

for eliciting requirements does not

have programming knowledge, they

miss many important requirements

[...] poorly deﬁned acceptance cri-

teria, when they exist.”

Business manage-

ment

”POs [Product Owners] that do not

actively participate in the process,

large companies, areas do not talk.”

”Delay in clients’ responses and

delay in receiving project details”

Midway changes in

requirements

”[...] requirements elicitation

starts in a way and, when de-

velopment starts, there are always

changes.”

”[...] difﬁculties caused by clients

when changing their minds along

the way”

”[...][poorly deﬁned requirements]

end up generating scope change in

the middle of development”

Requirements impos-

sible to be techni-

cally implemented

”[...] is a big challenge, along with

the limitations that we ﬁnd on plat-

forms”

”[...]Lack of test masks provided by

the client. Lack of APIs needed for

the feature the client wants to im-

plement in the bot.”

Lack of tools for re-

quirements engineer-

ing

”Lack of proper documentation

tools [...]”

Going from require-

ments to chatbots

”Create the context in which the bot

would ﬁt and create characters for

it.”

used documentation technique. Although this is not

alarming considering the nature of chatbots, it is es-

sentially a new technique in requirements engineer-

ing. Also, whereas use cases and scenarios are pop-

ular in other agile contexts (Jarzbowicz and Połocka,

2017; Jarzbowicz and Sitko, 2019), they are not very

used for chatbots.

The fact that chatbot requirements are often doc-

umented as conversation ﬂows may justify why re-

spondents used a lot of tools in requirements elicita-

tion and documentation since there is no established

way of documenting conversation ﬂows besides pro-

totyping them. Tools cited by practitioners are much

more related to elicitation than documentation. This

fact is supported by the responses of Q8, in which

most of the practitioners afﬁrmed the need to adapt

practices to elicit and document chatbot requirements.

Using inadequate tools for documenting requirements

can especially impair the traceability of requirements,

which are critical for the transparency of AI systems

(Mora-Cantallops et al., 2021).

The importance of proper documentation tools

for requirements is reinforced by one respondent in

the open ﬁeld when questioned about the challenges

of chatbot requirements elicitation and documenta-

tion. This importance is inferred from the responses

of three respondents that have suffered with midway

changes in requirements which could be more man-

ageable if there were proper tools to keep require-

ments traceability, especially in the case of docu-

menting them as conversation ﬂows. Moreover, the

challenges reported by respondents focus heavily on

human processes and information exchange, which

are not very distant from broader contexts, such as

agile development. The key challenges in agile re-

quirements engineering are technical dependencies

among teams; coordinate decisions; not lose sight of

the big picture; continuous management of require-

ments; work out requirements in cooperation with

(end users), and involve stakeholders throughout the

whole development process (Sch

on et al., 2017).

According to our respondents, the biggest chal-

lenge in eliciting requirements is meeting stake-

holder’s needs. This is a challenge in software de-

velopment in general, but it is an even bigger prob-

lem in chatbot development since software qualities

in conversation agents are much harder to achieve.

Traditional software has to guarantee functionality,

usability, reliability, performance, and supportability,

whereas chatbots have to guarantee these quality char-

acteristics plus proactivity, conscientiousness, com-

municability, emotional intelligence, (Chaves and

Gerosa, 2021) and other human attributes which are

harder to achieve. Lastly, the dispersion in responses

Requirements Engineering Challenges and Techniques in Building Chatbots

185

of Q7, Q9, Q10, and Q11 show that requirements gen-

erally receive different treatment and are given differ-

ent importance depending on the team working in it.

However, since responses are more concentrated in

the negative spectrum of options, they reveal that re-

quirements are probably being neglected by chatbot

teams, especially non-functional requirements.

6 THREATS TO VALIDITY

The internal validity of this work is threatened mostly

by online-based and unsupervised nature of our sur-

vey since we did not have control over the participants

when they answering our survey, although we made

our contact available for them in case there were any

doubts. On the other hand, participants were not pres-

sured to respond by being unsupervised and if they

were not compromised in responding the survey they

could quit at any time, which makes us believe that

the ones that ﬁnished responding provided meaning-

ful responses. The external validity is threatened by

fact that our survey included only Brazilian practi-

tioners and we could not guarantee that we shared it

with professionals of all contexts. Nevertheless, we

did reach professionals from many companies and re-

gions from the country albeit they were not obligated

to respond the survey since it was only an invitation

and we guarantee anonymity in the survey.

7 CONCLUSION

In this work, we surveyed chatbot practitioners to dis-

cover what are the techniques and tools they have

used and the challenges they face in chatbot require-

ments elicitation and documentation. We also ques-

tioned them about the relevance of requirements in

their projects. Our main contribution is to reveal difﬁ-

culties and the need for adaptation in requirements en-

gineering processes to be addressed in future works.

After gathering results we compared them with

the techniques and challenges faced in other software

development contexts. In general, the top used tech-

niques are the same used in other software ﬁelds.

However, interviews are the most used method in

other contexts, whereas brainstorming is the most

used method by chatbot practitioners. In the context

of our research, the fact that stands out is that con-

versation ﬂows have emerged as a new way of docu-

menting requirements and practitioners are negatively

impacted by the lack of tools for this speciﬁc purpose.

Lastly, the challenges faced in eliciting and doc-

umenting requirements are heavily related to require-

ments management and information exchange. Also,

responses indicate that meeting stakeholders’ needs

is one of the biggest challenges and we suppose that

is due to the fact that chatbots have quality attributes

harder to achieve since they are subjective human

characteristics.

REFERENCES

Adnan, S. M., Hamdan, A., and Alareeni, B. (2020). Arti-

ﬁcial intelligence for public sector: chatbots as a cus-

tomer service representative. In International Con-

ference on Business and Technology, pages 164–173.

Springer.

Akkiraju, R., Sinha, V., Xu, A., Mahmud, J., Gundecha, P.,

Liu, Z., Liu, X., and Schumacher, J. (2020). Char-

acterizing machine learning processes: a maturity

framework. In International Conference on Business

Process Management, pages 17–31. Springer.

Alﬂen, N. and Prado, E. V. (2021). Requirements elic-

itation techniques for software development: a sys-

tematic review of literature. AtoZ: novas pr

aticas em

informac¸

ao e conhecimento, 10(1):39–49.

Amazon (2021). Designing for conversation.

Belani, H., Vukovic, M., and Car, Z. (2019). Requirements

engineering challenges in building ai-based complex

systems. In 2019 IEEE 27th International Require-

ments Engineering Conference Workshops (REW),

pages 252–255.

Brandtzaeg, P. B. and Følstad, A. (2017). Why people use

chatbots. In International conference on internet sci-

ence, pages 377–392. Springer.

Brandtzaeg, P. B. and Følstad, A. (2018). Chatbots:

changing user needs and motivations. Interactions,

25(5):38–43.

Chaves, A. P. and Gerosa, M. A. (2021). How should my

chatbot interact? a survey on social characteristics

in human–chatbot interaction design. International

Journal of Human–Computer Interaction, 37(8):729–

758.

Dar, H., Lali, M. I., Ashraf, H., Ramzan, M., Amjad, T.,

and Shahzad, B. (2018). A systematic study on soft-

ware requirements elicitation techniques and its chal-

lenges in mobile application development. IEEE Ac-

cess, 6:63859–63867.

Fadhil, A. (2018). Can a chatbot determine my diet?: Ad-

dressing challenges of chatbot application for meal

recommendation. arXiv preprint arXiv:1802.09100.

Følstad, A. and Brandtzæg, P. B. (2017). Chatbots and the

new world of hci. interactions, 24(4):38–42.

Følstad, A. and Brandtzaeg, P. B. (2020). Users’ expe-

riences with chatbots: ﬁndings from a questionnaire

study. Quality and User Experience, 5(1):1–14.

Franch, X., Mendez, D., Vogelsang, A., Heldal, R., Knauss,

E., Oriol, M., Travassos, G., Carver, J. C., and Zim-

mermann, T. (2020). How do practitioners perceive

the relevance of requirements engineering research?

ICAART 2022 - 14th International Conference on Agents and Artiﬁcial Intelligence

186

IEEE Transactions on Software Engineering, pages 1–

Gobov, D. and Huchenko, I. (2020). Requirement elic-

itation techniques for software projects in ukrainian

it: An exploratory study. In 2020 15th Conference

on Computer Science and Information Systems (Fed-

CSIS), pages 673–681.

Google (2021). Conversation design.

Habibullah, K. M. and Horkoff, J. (2021). Non-functional

requirements for machine learning: Understanding

current use and challenges in industry. arXiv preprint

arXiv:2109.00872.

Harwell, D. (2018). The accent gap. The Washington Post.

IBM (2021). Conversational ux design.

Ishikawa, F. and Matsuno, Y. (2020). Evidence-driven

requirements engineering for uncertainty of machine

learning-based systems. In 2020 IEEE 28th Inter-

national Requirements Engineering Conference (RE),

pages 346–351.

Jakic, A., Wagner, M. O., and Meyer, A. (2017). The im-

pact of language style accommodation during social

media interactions on brand trust. Journal of Service

Management.

Jarzbowicz, A. and Połocka, K. (2017). Selecting re-

quirements documentation techniques for software

projects: A survey study. In 2017 Federated Confer-

ence on Computer Science and Information Systems

(FedCSIS), pages 1189–1198.

Jarzbowicz, A. and Sitko, N. (2019). Communication and

documentation practices in agile requirements engi-

neering: A survey in polish software industry. In Eu-

roSymposium on Systems Analysis and Design, pages

147–158. Springer.

Khan, R. and Das, A. (2018). Introduction to chatbots. In

Build Better Chatbots: A Complete Guide to Getting

Started with Chatbots, pages 1–11. Apress, Berkeley,

CA.

Kostova, B., Gurses, S., and Wegmann, A. (2020). On the

interplay between requirements, engineering, and ar-

tiﬁcial intelligence. In REFSQ Workshops.

Lim, T.-Y., Chua, F.-F., and Tajuddin, B. B. (2018). Elic-

itation techniques for internet of things applications

requirements: A systematic review. In Proceedings

of the 2018 VII International Conference on Network,

Communication and Computing, pages 182–188.

Mendonc¸a, G. D., de Souza Filho, I. P., and Guedes, G. T. A.

(2021). A systematic review about requirements engi-

neering processes for multi-agent systems. In ICAART

(1), pages 69–79.

Mora-Cantallops, M., S

anchez-Alonso, S., Garc

ıa-

Barriocanal, E., and Sicilia, M.-A. (2021). Traceabil-

ity for trustworthy ai: A review of models and tools.

Big Data and Cognitive Computing, 5(2):20.

Neff, G. and Nagy, P. (2016). Automation, algorithms, and

politics— talking to bots: Symbiotic agency and the

case of tay. International Journal of Communication,

10:17.

Petriv, Y., Erlenheim, R., Tsap, V., Pappel, I., and Draheim,

D. (2019). Designing effective chatbot solutions for

the public sector: A case study from ukraine. In In-

ternational Conference on Electronic Governance and

Open Society: Challenges in Eurasia, pages 320–335.

Springer.

Pressman, R. (2009). Software Engineering: A Practi-

tioner’s Approach. McGraw-Hill, Inc., USA, 7 edi-

tion.

Raﬁq, U., Bajwa, S. S., Wang, X., and Lunesu, I. (2017).

Requirements elicitation techniques applied in soft-

ware startups. In 2017 43rd Euromicro Conference

on Software Engineering and Advanced Applications

(SEAA), pages 141–144. IEEE.

Sch

on, E.-M., Winter, D., Escalona, M. J., and

Thomaschewski, J. (2017). Key challenges in agile

requirements engineering. In International Confer-

ence on Agile Software Development, pages 37–51.

Springer, Cham.

Shevat, A. (2017). Designing bots: Creating conversational

experiences. ”O’Reilly Media, Inc.”.

Singh, A., Ramasubramanian, K., and Shivam, S. (2019).

Identifying the sources of data. In Building an En-

terprise Chatbot: Work with Protected Enterprise

Data Using Open Source Frameworks, pages 19–33.

Apress, Berkeley, CA.

Sommerville, I. (2010). Software Engineering. Addison-

Wesley Publishing Company, USA, 9th edition.

Soni, N., Sharma, E. K., Singh, N., and Kapoor, A.

(2019). Impact of artiﬁcial intelligence on busi-

nesses: from research, innovation, market deployment

to future shifts in business models. arXiv preprint

arXiv:1905.02092.

Sun, T. Q. and Medaglia, R. (2019). Mapping the chal-

lenges of artiﬁcial intelligence in the public sector:

Evidence from public healthcare. Government Infor-

mation Quarterly, 36(2):368–383.

Vogelsang, A. and Borg, M. (2019). Requirements engi-

neering for machine learning: Perspectives from data

scientists. In 2019 IEEE 27th International Require-

ments Engineering Conference Workshops (REW),

pages 245–251. IEEE.

Wautelet, Y., Heng, S., Kolp, M., and Scharff, C. (2016).

Towards an agent-driven software architecture aligned

with user stories. In ICAART (2), pages 337–345.

Wilhjelm, C. and Younis, A. A. (2020). A threat analysis

methodology for security requirements elicitation in

machine learning based systems. In 2020 IEEE 20th

International Conference on Software Quality, Relia-

bility and Security Companion (QRS-C), pages 426–

433.

Xu, Y., Shieh, C.-H., van Esch, P., and Ling, I.-L. (2020). Ai

customer service: Task complexity, problem-solving

ability, and usage intention. Australasian Marketing

Journal (AMJ), 28(4):189–199.

Zhang, J. J., Følstad, A., and Bjørkli, C. A. (2021). Orga-

nizational factors affecting successful implementation

of chatbots for customer service. Journal of Internet

Commerce, pages 1–35.

Requirements Engineering Challenges and Techniques in Building Chatbots

187