Professional Documents
Culture Documents
Kcomt Lopez Wong2022
Kcomt Lopez Wong2022
using a chatbot
Luis Kcomt Lam Cesar Andres Lopez Hurtado Lenis Wong Portillo
Faculty of Engineering Faculty of Engineering Faculty of Engineering
Universidad Peruana de Ciencias Universidad Peruana de Ciencias Universidad Peruana de Ciencias
Aplicadas Aplicadas Aplicadas
Lima, Peru Lima, Peru Lima, Peru
0000-0002-8832-8929 0000-0001-9716-0155 0000-0002-5032-3233
Abstract— Requirement elicitation determines the success of implementing a system that offers automated US extraction
a project given that it describes the needs of the context and the via a structured interview, between users and a chatbot, and
problem at hand. However, a recent Chaos Report showed that other software management tools for correctly handling
projects are increasingly failing in large numbers. This is software requirements. We rely on IWA for the identification
generally due to the poor quality of requirements that are a of intents and entities, and we implement an algorithm for the
result of an erroneous requirement elicitation phase (REP). This extraction of US based on the identified elements.
affects the scalability of the REP in complex environments. In
this paper, we propose a framework to automate the REP, II. RELATED WORK
validated by a system that consumes a chatbot service from IBM
Watson Assistant (IWA). The validation is done by comparing Regarding the related work, the methodology proposed by
traditional requirement elicitation with requirement elicitation [2] was used to obtain and analyze several articles that were
using our tool with the help of a Peruvian software company. aligned with the objectives of our research. The methodology
Finally, we demonstrate that requirement elicitation with our consists of 3 phases: planning, research and results, and
implemented tool is more efficient than traditional requirement analysis.
elicitation and results in higher quality requirements.
First, for the planning phase, 4 research questions were
Keywords— User Stories; IBM Watson Assistant; posed that served as objectives: RQ1: What aspects have been
Requirement elicitation; Chatbot touched in the automation of RE? RQ2: What techniques have
been applied in RE? RQ3: What natural language processing
I. INTRODUCTION approaches have been applied to integrate chatbots with
requirement elicitation? and RQ4: What factors affect the
Software requirements are characteristics of a system
requirements REP?
expressed by the stakeholders to delimit its functionality.
These requirements are captured during a process called the Second, for the research phase, a search was done to find
REP [1]-[3]. This process is traditionally carried out between related studies that answered at least one of the research
users and analysts and determines the success of the later questions. The search was done manually, using the following
phases of a products life cycle [4] [5]. However, this phase is keywords: "natural language processing", "requirement
being carried out erroneously. This can be seen in the 2019 elicitation", "chatbot", "requirement engineering",
CHAOS Report [6], where it was estimated that around 52.7% "automation in requirement engineering". It was done through
of all projects end with failures, and 31.1% fail; that is, only search engines such as Scopus, IEEE, Science Direct, etc.
about 16.2% of all projects are completely successful. The Mostly, only articles that had been published in the last 3 years
Chaos report considered these failures mainly a consequence and had a high level of impact (Q1, Q2) were chosen.
of lack of resources, lack of planning and poor-quality
requirements [7]. Third, for the results phase, 10 papers were chosen and
analyzed based on the exclusion criteria mentioned
Currently, there are some tools that help improve how previously. The analysis of the papers was done by classifying
analysts carry out the REP, nevertheless, most are oriented these in different taxonomies created based on the research
towards users with medium to high levels of experience [8]. questions posed (see Table I).
Therefore, the REP cannot be applied in large or complex
contexts, since it implies involving novice users in the REP, TABLE I. TAXONOMY OF COLLECTED PAPERS
resulting in poor quality requirements [2]. Taxonomy References
Regarding recent approaches in the field of chatbots and Aspects (RQ1) [1], [5], [8], [13]
Techniques (RQ2) [4], [5], [8], [9], [10], [14], [13]
requirement engineering (RE). There are various methods for
Approaches (RQ3) [1], [3], [5], [8]
extracting functional requirements (FR) and non-functional Factors (RQ4) [5], [11], [13]
requirements (NFR). However, few result in user stories (US)
as such, most provide matrices, trees, or attributes,
Regarding the classification “aspects” touched in
consequences, and core values (ACV) chains [10]. In addition,
automation, the following results were obtained: automatic
few studies used chatbots to capture requirements.
elicitation [1], cloud computing [13], automatic extraction [4]
This study is motivated by the impact that REP has on the and automatic correction [14]. Automatic extraction mainly
success of a software project [4], and the difficulty it has when contained techniques that extracted the NFR based on
scaling [8], due to lack of novice-oriented tools. Therefore, we documents or other FR, whereas automatic elicitation is
propose a framework for automating the REP via the use of a based on capturing the requirements automatically in real
chatbot provided by IWA. We validate our framework by time.