Generating Process Models by Interacting with Chatbots—A Literature Review

Abstract

This paper presents a comprehensive literature review, focusing on the emerging intersection of chatbot technology and the generation of process models. As an evolving field of study, the integration of interactive chatbots into process model generation represents a promising approach, blending advancements in artificial intelligence in general, and natural language processing in particular, with process management methods. This review systematically examines the existing literature across multiple disciplines, identifying and analyzing studies that touch upon the individual components of this nascent topic: chatbot technology, process model generation, and their synergistic potential. Despite the scarcity of direct research aimed at using chatbots for process model generation, this review synthesizes relevant findings from related domains, such as natural language processing applications in process modeling, and the broader impact of chatbot interfaces in various domains. Through this analysis, we aim to map the current landscape of research, highlight significant gaps, and suggest potential pathways for future investigations. This paper not only aggregates existing knowledge, but also assesses the applicability and implications of current technologies and theories when generating process models with the assistance of interactive chatbots. The outcome is a foundational compendium for researchers and practitioners interested in exploring this innovative intersection, providing a springboard for future research and development in this promising area.

1. Introduction

Process Modeling has been a fundamental component in the design and optimization of information systems for a long time, providing a framework for understanding and enhancing organizational processes. In the current era of digital transformation, process modeling faces the challenge of adapting to rapidly evolving technological landscapes [1]. The integration of chatbots, empowered by advancements in artificial intelligence (AI) and Natural Language Processing (NLP), particularly large language models, has the potential to revolutionize process modeling by enabling more intuitive and interactive ways of modeling business processes [2]. Interactive chatbots, equipped with the capability to understand and interpret human language, offer an intuitive and efficient means for various stakeholders to engage with process modeling. Chatbots can potentially streamline data collection, enable model updates, and facilitate user-friendly interactions (e.g., querying information from process models). This shift not only promises to make process modeling more accessible, but also aims to enhance the accuracy and relevance of models by capturing real-time inputs and diverse user perspectives. The relevance of this technology extends beyond simplifying the technical aspects of process modeling. In a broader context, it has the potential to significantly impact various fields, such as business process management, information systems, and organizational development. By automating and optimizing process model generation, organizations might achieve greater agility, better compliance, and enhanced operational efficiency. Furthermore, the democratization of process modeling through chatbots—meaning the creation of tools and techniques making the process of process modeling more accessible to a broader range of people, regardless of their technical expertise—can lead to more inclusive and collaborative workplace environments.

However, the integration of chatbots in process model generation comes with several challenges. Concerns regarding the accuracy of chatbot-generated models and issues of data security and privacy are paramount. In additions, concerns regarding the quality, correctness, and better comprehensibility of the generated process models, as well as a better collaboration between business process analysts and domain experts, are crucial. Through a comprehensive literature review, this paper investigates the current state of research in this domain, identifies gaps in the knowledge, and proposes pathways for future investigations. The aim is to provide a foundational understanding of the capabilities and limitations of integrating chatbots for generating process models, setting the stage for its evolution and effective implementation in various fields.

The paper is structured as follows: In Section 2, the background and methodology of the systematic literature review (SLR) is described. Section 3 discusses the research questions we are interested in and the search strategy we adopted. In Section 4, the execution of the SLR is explained and an overview of the SLR results is presented. Section 5 presents the studies, followed by a discussion of the results in Section 6. Finally, Section 7 discusses threads to validity and Section 8 concludes the paper.

2. Background and Methodology

This section provides essential definitions that serve as the foundation for the SLR. The terms we introduce are frequently referenced throughout the paper to ensure clarity and consistency of our analysis. Following the definitions, we outline the methodology of the SLR, providing insights into the systematic approach employed in our research.

2.1. Background

Process model. A process model is a formalized representation of a business process, capturing the sequence of tasks, activities, or events that need to occur to achieve a specific process outcome. These models are typically used to understand, analyze, and improve business processes. They often include elements such as tasks, decision points, resources (e.g., actors or systems involved in the process), and the flow of information or materials between different tasks. Process models can be visualized through various diagramming techniques like flowcharts, Business Process Model and Notation (BPMN), or Unified Modeling Language (UML) activity diagrams [3].

Natural Language Processing (NLP). NLP is a branch of AI that focuses on the interaction between computers and human languages. It encompasses techniques for processing, analyzing, and understanding natural language text or speech data, enabling machines to interpret and generate human-like language [4].

Large Language Models (LLM). LLMs are advanced NLP models that utilize deep learning techniques to process and generate natural language text on a large scale. These models, trained on vast amounts of textual data, demonstrate remarkable capabilities in understanding and producing human-like language, revolutionizing various NLP tasks such as text generation, translation, and sentiment analysis [5].

Chatbot. A chatbot (also known as a chat robot) is an AI-powered software application designed to simulate conversation with human users via textual or auditory interfaces. Chatbots leverage NLP techniques, including LLMs, to interpret user input, generate responses, and engage in interactive dialogues across a wide range of domains, from customer service to personal assistance [6].

Prompting. Prompting is the act of presenting prompts or cues to an LLM during a conversation with a chatbot. These prompts may take various forms, such as questions, suggestions, or instructions, and are aimed at guiding the flow of dialogue and steering the LLM towards desired outcomes or goals [7].

2.2. SLR Methodology

A structured and systematic approach was employed for selecting and sourcing the literature. This methodology is pivotal to ensure that the review comprehensively encompasses the relevant studies, upholding both the breadth and depth of the research topic. The SLR was executed following the rigorously defined protocol by Kitchenham et al. [8], which is well-established in the software engineering and information systems fields. This protocol delineates a tripartite methodology, each phase unfolding sequentially and characterized as follows.

2.2.1. Planning the SLR

The planning phase encompasses the establishment of a robust review protocol. Critical activities in this phase involve delineating the SLR’s objectives with precision, devising a query for database interrogation, selecting an array of pertinent digital libraries for research retrieval, and meticulously defining inclusion and exclusion criteria to effectively sieve through the amassed studies.

2.2.2. Conducting the SLR

This phase is dedicated to the pragmatic implementation of the search strategy and the subsequent analytical evaluation of the retrieved scholarly works. The procedure commences with the deployment of the crafted query across the selected digital repositories, followed by the aggregation of the resultant academic articles. Each article is rigorously scrutinized, with the pre-established inclusion and exclusion criteria applied to distill a primary collection of relevant studies. Augmentation techniques such as backward and forward snowballing [9,10] are employed to enrich the corpus. Backward snowballing entails an examination of each study’s citations to unearth additional pertinent works, whereas forward snowballing involves identifying studies that reference the initially selected relevant articles [9]. These techniques are subject to the same stringent inclusion and exclusion criteria. The culmination of this phase is characterized by the extraction of data from the curated body of literature, facilitating the categorization and thematic clustering of the articles, and organizing the extracted data to coherently address the predefined research questions.

2.2.3. Reporting SLR Results

The concluding phase involves the meticulous composition and presentation of the review findings. This encompasses the synthesis and coherent articulation of the results, with a focus on elucidating the responses to the initially formulated research questions.

3. Structure of the Systematic Literature Review

This section presents the structure and design of the SLR, including the research questions, search strategy, search string, used digital libraries, and inclusion as well as exclusion criteria.

3.1. Research Questions

The overarching research question was formulated as ‘How can natural language and process models be integrated, including the use of chatbots?’. Starting with this research question, we conducted a SLR and subsequently divided this primary research question into four detailed research questions. The latter were obtained after having gained an overview of the collected studies, which we categorized according to their method, distinguishing between (1) studies generating process models from natural text input, (2) studies generating text from modeled process models, (3) studies encompassing both directions, and (4) studies that investigate the use of chatbots in the context of (1)–(3).

• RQ1: How to generate process models from natural language text input (Text-to-Model)? Research question RQ1 aims to explore and understand the methodologies, techniques, and approaches utilized in transforming unstructured natural language text into process models (cf. Figure 1). Through this investigation, the goal is to identify the various methods employed, and any emerging trends or advancements in the field.
• RQ2: How to generate natural language text from process models (Model-to-Text)? The aim of RQ2 is to investigate and comprehend the methodologies, techniques, and strategies employed in generating natural language text output from process models (cf. Figure 2). Through this inquiry, the objective is to discern the diverse approaches utilized.
• RQ3: How to get from natural language text input to process models and backwards? The aim of RQ3 is to explore methodologies, techniques, and frameworks that enable bidirectional conversions between natural language text and process models. By investigating this bidirectional transformation process, the goal is to understand the various approaches utilized.
• RQ4: How are chatbots included in the processes of RQ1–RQ3? The aim of RQ4 is to explore which of the collected studies for RQ1–RQ3 use a chatbot for generating process models from natural language text or for generating natural language text from process models.

3.2. Search Strategy

The focus of any SLR is primarily directed towards four key elements: (1) the formulation of the search query, (2) the selection of digital libraries, (3) the establishment of criteria for the inclusion or exclusion of research studies during the analysis, and (4) the implementation of the snowballing technique [9]. Each of these components plays a crucial role in ensuring a comprehensive and systematic approach to literature review, thereby facilitating a thorough exploration and synthesis of relevant academic insights.

3.2.1. Search String

A carefully curated set of keywords and a search term were employed to effectively navigate through specific digital libraries, which will be mentioned in more detail later. Two main areas were defined on which our topic of this SLR refers to process models and chatbots. Commencing from the identified thematic areas, a systematic process was undertaken to determine the pertinent terminologies to be incorporated into the search query.

The selection of the keywords for the search string involved a collaborative and iterative brainstorming process involving six experts. Three of them were from industry, working for companies that develop and distribute process modeling tools and Business Process Management (BPM) systems, respectively. Hence, they already have faced the challenge of integrating their tools with bot capabilities. Moreover, three experts and researchers from academia were involved that work in the fields of BPM and model-driven software engineering, respectively. This iterative discourse entailed multiple rounds of deliberation and refinement, culminating in a consensus on the final compilation of terms.

These terms, which form the foundation of this search strategy, are comprehensively enumerated in

Table 1. Search string keywords by domain.

Process modeling	(‘process model*’ OR ‘workflow*’ OR ‘BPMN’)
Chatbot	AND (‘Chatbot*’ OR ‘natural language processing’ OR ‘NLP’
	OR ‘GPT*’ OR ‘large language model*’ OR ‘LLM’)

, and are further explicated in the subsequent discourse. This meticulous approach ensures a precise and relevant query formulation, pivotal for the efficacy of the literature search.

The phrases process model*, workflow* and BPMN were used to refer to the field of business process management. Since process modeling is the subject of this research, process model* and BPMN were included to refer to the entire field of this SLR. The term process model*, which refers to the general concept of modeling processes, was added retrieve literature, which refers to a wider range of processes. Finally, with workflow, the Workflow Management Coalition refers to a business process [11]; therefore, the keyword workflow has been added as well. Within the realm of chatbot, the following keywords were added:chatbot*, natural language processing, NLP, GPT*, large language model*, and LLM. Since chatbots constitute the focus of this research in addition to the already-discussed keyword process model*, in light of the prominence of advanced NLP technologies, as exemplified by the chatbot application ChatGPT, keywords associated with Large Language Model* and Natural Language Processing were also incorporated into our search criteria.

Subsequent to the selection of relevant terms, the construction of the search string was methodically undertaken. This involved employing the logical disjunction connective (‘OR‘) to amalgamate terms within the same thematic area, while utilizing the logical conjunction connective (‘AND’) to integrate the sets of terms across diverse thematic areas. This systematic approach in query formulation ensured a comprehensive and precise search strategy. (It is important to note that the asterisk (*) character serves to broaden the search criteria by including any word that begins with the characters preceding it. For instance, specifying ‘process model*’ enables us to encompass variations like ‘process models’ in our search results).

3.2.2. Digital Libraries

Upon finalizing the query parameters, we proceeded to select the pertinent digital libraries for this research. These databases were chosen due to their broad coverage across the fields of Computer Science, AI, and Business Process Management, ensuring a diverse and comprehensive collection of literature relevant to the research topic. The selection prioritized the most prominent digital libraries in the fields of academic and professional publications in the above fields: IEEE Explore, ScienceDirect, Springer Link, Web of Science, and ACM Digital Library. Furthermore, we specified the search within these digital libraries, confining the query to titles, abstracts, and keywords. This criterion implies that any study featuring a conjunction of terms from the domains of process modeling and chatbots, whether in the title, abstract, or keywords, were included in our query results. Nevertheless, the search was adapted based on the database because it was only possible in some databases to limit search fields (e.g., restricting the query to titles, abstracts, and keywords).

3.2.3. Inclusion and Exclusion Criteria

The inclusion criteria for the literature were precisely defined to ensure relevance, quality, and comprehensiveness. Firstly, studies that directly addressed the integration of chatbots or LLMs in process modeling were prioritized, as this is the primary focus of this SLR. The scholarly merit of the sources was a non-negotiable criterion, with an emphasis on peer-reviewed journals, conferences, and academic proceedings to maintain academic rigor. Conversely, the exclusion criteria were established to maintain the academic integrity and focus of the review. Non-academic sources such as magazines, blogs, and informal publications were excluded. Literature that was overly focused on unrelated technological applications or did not contribute directly to the discussion of process modeling or AI-driven tools was also excluded.

Additionally, no temporal boundary was set for the query. Literature searches were conducted without specifying a starting year, allowing for the inclusion of both foundational studies and recent research up to the year 2024. This approach was adopted to capture the evolution of the field, ensuring the analysis included both early conceptual frameworks and the latest stabilized data relevant to today’s technological advancements.

4. Conducting the Systematic Literature Review

This section delineates the methodological framework underpinning the conducting of the SLR. Specifically, it elucidates the procedural nuances associated with the identification, selection, and reporting of the studies. Furthermore, this section also delves into the groups employed for categorizing of the analyzed literature.

4.1. Identification and Selection of the Studies

Ending in January 2024, following the implementation of the search protocol, a total of 18 relevant studies were identified. The BPMN process model, depicted in Figure 3, shows the sequence of selection steps we employed, with (n) $\widehat{=}$ amount of studies.

The initial application of the search query across the selected digital libraries yielded a collection of 1148 potentially relevant studies. This pool included contributions from IEEE Explore (336), ScienceDirect (313), Springer Link (14), Web of Science (444), and ACM Digital Library (41). These databases were meticulously chosen for their extensive coverage in the realms of computer science, AI, and business process management, thereby ensuring a thorough and varied assemblage of scholarly literature pertinent to the study.

Subsequently, we conducted a review of the titles of the 1148 studies. The aim of this phase was to ascertain the suitability of these works for the described research, particularly by applying the presented exclusion criteria (cf. Section 3.2.3). We also removed duplicates.

Afterwards, we conducted a more detailed review of the titles and abstracts. A significant number of studies were discarded at this stage, as it became evident from their titles and abstracts that they did not encompass approaches related to chatbots generating process models. This refinement process resulted in a narrowed set of 16 relevant works. These 16 studies were read in full, and one of them was then determined to not fit our considered works.

To enhance the precision of this review, and to validate the studies selected, we employed the snowballing technique [9], beginning with the 15 studies that remained. The snowballing process, capable of multiple iterations, was reapplied to newly discovered relevant studies in each subsequent iteration. During the first iteration of snowballing, three studies were added. During the second iteration, no new study was found to be added to our collection. The process was halted after the first iteration, as it only yielded duplicate studies, indicating that the saturation point of relevant literature had been reached.

4.2. Data Extraction and Synthesis

In the context of this SLR, the data extraction and synthesis phases were carefully conceived to facilitate the systematic recording and aggregation of pertinent information derived from the analysis of the selected studies.

Table 2. Relevant studies in this SLR.

Study #	Title	Reference	Author
S1	Extracting process models using NLP Technique	[12]	Konstantinos Sintoris, Kostas Vergidis
S2	A Proposed Technique for process modeling Diagram using Natural Language Processing	[13]	Shorouq Elmanaseer, Ahmad AA Alkhatib, Rand N Albustanji
S3	A combined model of NLP with process modeling for Sentiment Analysis	[14]	Dhivyashree M, Sarum A Thi K R, SabarmA Thi K R
S4	Comparing Generative Chatbots Based on Process Requirements: A Case Study	[15]	Luis Fernando Lins, Nathalia Nascimento, Paulo Alencar, Toacy Oliveira, Donald Cowan
S5	Assisted Declarative Process Creation from Natural Language Descriptions	[16]	Hugo A.Lopez, Morten Marquard, Lukas Muttenthaler, Rasmus Stromsted
S6	A Case Study on Designing Business Processes Based on Collaborative and Mining Approaches	[17]	João Carlos de A.R. Gonçalves, Flávia Maria Santoro, Fernanda Araujo Baião
S7	Quantifying chatbots’ ability to learn business processes	[18]	Christoph Kecht, Andreas Egger, Wolfang Kratsch, Maximilian Röglinger
S8	Generating BPMN diagram from textual requirements	[19]	Sholiq Sholiq, Riyanarto Sarno, Endang Siti Astuti
S9	A visual approach for identification and annotation of business process elements in process descriptions	[20]	Leonardo Silva Rosa, Thanner Soares Silva, Marcelo Fantinato, Lucineia Heloisa Thom
S10	Extracting Declarative Process Models from Natural Language	[21]	Han van der Aa, Claudio Di Ciccio, Henrik Leopold, Hajo A. Reijers
S11	A Semi-automatic Approach to Identify Business Process Elements in Natural Language Texts	[22]	Renato Cesar Borges Ferreira, Lucineia Heloisa Thom, Marcelo Fantinato
S12	Supporting the Process of Learning and Teaching Process Models	[23]	Josep Sanchez-Ferreres, Luis Delicado, Amine Abbad Andaloussi, Andrea Burattin, Guillermo Calderon-Ruiz, Barbara Weber, Josef Carmona, Lluıs Padro
S13	Automatic Process Model Discovery from Textual Methodologies	[24]	Elena Viorica Epure, Patricia Martín-Rodilla, Charlotte Hug, Rebecca Deneckère, Camille Salinesi
S14	Live process modeling with the BPMN Sketch Miner	[25]	Ana Ivanchikj, Souhaila Serbout, Cesare Pautasso
S15	Process Model Generation from Natural Language Text	[26]	Fabian Friedrich, Jan Mendling, Frank Puhlmann
S16	A Concept for Generating process models from Natural Language Description	[27]	Krzysztof Honkisz, Krzysztof Kluza, Piotr Wiśniewski
S17	A Machine Translation Like Approach to Generate process model from Textual Description	[28]	Riad Sonbol, Ghaida Rebdawi and Nada Ghneim
S18	Natural language processing with process models	[29]	Jan Mendling, Henrik Leopold, Lucineia Heloisa Thom, Han van der Aa

,

Table 3. Content of the relevant Studies S1–S8 in this SLR.

Study	Content
S1	S1 discusses the application of NLP to automatically generate process models from existing documentation within organizations. It explores how syntactic and grammatical structures of sentences can be utilized to identify process model elements such as activities, resources, and tasks. This approach aims to streamline the process modeling phase by leveraging automated tools to interpret and convert natural language documentation into structured process models.
S2	S2 describes an approach that integrates NLP to facilitate the creation of process models. This technique leverages probabilistic latent semantic analysis during the preprocessing stage to manage the complexity of natural language with multiple meanings. The method aims to enhance the accuracy and efficiency of generating process models directly from textual descriptions, significantly streamlining the modeling process.
S3	S3 explores the integration of NLP techniques with business process modeling to enhance sentiment analysis capabilities. This approach utilizes NLP to analyze textual data for sentiment and incorporates business process models to contextualize and improve the accuracy of sentiment detection. Study S3 suggests that combining these methodologies can provide more nuanced insights into sentiment data, which is particularly beneficial for applications like monitoring customer feedback and optimizing business processes.
S4	S4 evaluates the effectiveness of generative chatbots, like ChatGPT, in managing and executing business processes according to predefined requirements. It systematically assesses these chatbots through evaluation questions categorized into six key areas, such as Start Event, Forward Flow, and End Event, which highlight their ability to adhere to BPMN standards. The findings of Study S4 suggest that while these chatbots perform well in straightforward scenarios, difficulties emerge for tasks requiring detailed decision-making, indicating a need for further enhancements in their processing capabilities.
S5	S5 explores advancements in helping users generate declarative process models from natural language texts. S5 introduces the Process Highlighter, a tool that assists users in manually creating Dynamic Response Condition (DCR) graphs, i.e., declarative process models, from textual documents. This facilitates non-technical users in adopting declarative process models, making it easier to convert complex text into structured process formats.
S6	S6 investigates how collaborative and mining techniques can enhance business process design. It presents a comprehensive analysis of the Ghana cocoa supply chain to highlight how information technology can address the asymmetry in information access among stakeholders, thereby supporting more inclusive business models. The case study details the transformation from existing business processes to more efficient and inclusive processes enabled by IT systems, aiming to improve livelihoods and economic output in the cocoa sector.
S7	S7 develops a method to measure how effectively chatbots can understand and execute business processes based on process mining metrics. S7 uses a large dataset from customer service interactions on Twitter (X) to train chatbots and to evaluate their performance against normative process models. Study S7 highlights the potential of chatbots in adhering to organizational procedures, thus aiding their broader adoption in practical scenarios.
S8	S8 explores a method for translating natural language descriptions into BPMN process models. The approach involves two main stages: analyzing textual requirements using NLP techniques to extract fact types and then using these fact types to generate BPMN diagrams. This method allows for the conversion of informal business process descriptions into formalized, visual BPMN diagrams, facilitating better understanding and communication of business processes.

and

Table 4. Content of the relevant Studies S9–S18 in this SLR.

Reference	Content
S9	S9 proposes a user-interactive visual methodology to identify and annotate BPMN 2.0 elements within textual process descriptions. The approach uses sequences of words to detect process elements, creates a consistent data structure, and makes it available as a web service. This methodology supports process comprehension and assists users in the process modeling phase by enhancing the accuracy of annotations through visual interaction.
S10	S10 presents an automated method for generating declarative process models from textual descriptions. This approach uses tailored NLP techniques to identify activities and their interrelations within textual constraints, facilitating the creation of process models that capture complex behaviors.
S11	Study S11 introduces a method that combines NLP and empirical mapping rules to semi-automatically identify business process elements from texts. This enhances the efficiency of process modeling by improving the accuracy of element extraction compared to traditional manual methods. The approach is validated through a proof of concept prototype, highlighting potential for further refinement and application in diverse business contexts.
S12	Study S12 presents a framework to aid both novice and experienced process modelers in creating accurate process models. The approach involves a model called ‘Model Judge,’ which evaluates the syntactical and semantic correctness of process models, providing corresponding feedback. This tool is designed to enhance the learning and teaching experience.
S13	S13 addresses the challenge of mining process models from unstructured textual data. The authors developed an unsupervised technique called TextProcessMiner, which identifies process activities and their relationships that uses NLP focusing on verb semantics.
S14	S14 describes a tool designed to help process modelers in quickly creating BPMN process models from natural language process descriptions. This tool supports live modeling during interviews and workshops, allowing participants to obtain immediate visual feedback and ensure the accuracy of the process representation. Usability and performance of the BPMN Sketch Miner was evaluated for usability and performance, showing that it can handle large models efficiently.
S15	S15 presents an innovative approach for automatically generating BPMN process models from natural language text. It integrates existing NLP tools and enhances them with an anaphora resolution mechanism to ensure coherent model generation. This methodology facilitates the transformation of informal process descriptions into formal BPMN models, making it easier to visualize and analyze business processes. The approach aims to bridge the gap between textual process documentation and executable process models.
S16	S16 presents a methodology for transforming natural language descriptions into process models. The proposed method involves the use of a spreadsheet-based intermediate model to capture the extracted information, which is then converted into a BPMN process model.
S17	S17 proposes a method that uses a semantic transfer-based machine translation approach to create process models from text. This approach includes two phases: (1) natural language analysis and (2) BPMN process model generation through semantic, syntactic, and morphological manipulations.
S18	Study S18 investigates integrating NLP techniques with process modeling to improve automatic extraction and understanding of process-related information from textual descriptions. Study S18 utilizes advanced NLP methods such as machine learning, semantic analysis, and syntactic parsing to identify and structure process elements within texts.

provide an overview of the content of all relevant studies identified by the SLR.

The formulation of attributes, encompassing Research Questions RQ1–RQ3, was an outcome of a comprehensive content analysis procedure. We thoroughly analyzed the studies, identifying the methods utilized for each research question, and established them as attributes. It is noteworthy that attributes ($A_i$) were defined for RQ1–RQ3, reflecting our initial focus on investigating the methods used to address these questions. Initially, however, attributes were not defined for RQ4. Instead, we first aimed to explore the methods employed for RQ1–RQ3. In RQ4, we solely contemplate which of the studies utilize a chatbot to address RQ1–RQ3. The established attributes for each research question RQ1–RQ3 are presented in

Table 5. Distribution of attributes for each Research Question.

Research Question	${\mathit{A}}_1$ (NLP)	${\mathit{A}}_2$ (Mapping Rules)	${\mathit{A}}_3$ (Sentence Level Analysis)
RQ1	√	√	√
RQ2	√
RQ3	√

, providing a structured overview of the data categorization framework employed in this research, followed by the definitions of the attributes.

To answer Research Question RQ1 (how to generate process models from natural language text input), in the following, we present the most frequently used technologies and methodologies of the studies to generate process models from natural language text input:

• RQ1$A_1$: NLP. NLP involves the computational analysis of human language to extract meaning and structure from unstructured text data. In the context of generating process models from natural language text input, NLP techniques are employed to parse and understand textual descriptions of processes. This may include tasks such as part-of-speech tagging, syntactic parsing, named entity recognition, and semantic analysis. By applying NLP algorithms, systems can identify key process elements, relationships, and activities described in the text, which can then be translated into formal representations such as BPMN diagrams.
• RQ1$A_2$: Mapping rules. Mapping rules refer to predefined guidelines or algorithms that are used to translate natural language descriptions into formal representations, such as BPMN process models. These rules define how specific linguistic constructs or patterns in text correspond to elements and relationships in the target model. Mapping rules may be based on syntactic analysis, semantic rules, domain-specific knowledge, or a combination thereof. By applying mapping rules systematically, systems can automate the process of generating process models from textual input, ensuring consistency and accuracy in the resulting models.
• RQ1$A_3$: Sentence-level analysis. Sentence-level analysis involves the examination and interpretation of individual sentences within a text to extract relevant information and structure. In the context of generating process models from natural language text as input, sentence-level analysis focuses on identifying process-related statements, actions, and dependencies expressed at the sentence level. This may involve techniques such as sentence parsing, sentiment analysis, and keyword extraction to recognize process activities, participants, and constraints. Sentence-level analysis enables finer-grained understanding of textual descriptions, facilitating the creation of detailed and contextually relevant process models.

To answer Research Question RQ2 (how to generate natural language text from process models), we present the most frequently used technology and methodology in the studies and discuss of how they generate natural language text from process models:

• RQ2$A_1$: NLP. In the context of generating natural language text from process models, NLP techniques are utilized to transform structured representations, such as BPMN diagrams, into human-readable textual descriptions. This process involves analyzing the elements and relationships encoded in the process model, such as activities, events, gateways, and flows, and generating coherent and grammatically correct sentences that convey the semantics of the model. NLP algorithms may employ syntactic and semantic parsing, template-based generation, and linguistic rules to ensure the accuracy and fluency of the generated text. By leveraging NLP, systems can automate the generation of textual documentation, instructions, or explanations from process models, facilitating communication and comprehension among stakeholders. Additionally, NLP-based approaches enable the dynamic generation of text tailored to specific audiences or contexts, enhancing the usability and accessibility of process documentation in diverse organizational settings.

To answer Research Question RQ3 (How to get from natural language text input to process models and backwards), the same attribute as in RQ1 and RQ2 were selected and used.

• RQ3$A_1$: NLP. (cf. RQ1$A_1$ and RQ2$A_1$)

As mentioned above, for RQ4, we solely contemplate which of the studies utilize a chatbot to address RQ1–RQ3.

5. Reporting of the Systematic Literature Review

This section serves as a comprehensive exposition dedicated to the reporting of the SLR conducted on the topic of generating process models with interactive chatbots. We present a broad analysis of the 18 studies we identified, and then shift the focus to the specific data that was gathered to respond to the research questions (cf. Section 4). This section endeavors to provide a meticulous examination of the extant literature corpus, elucidating the methodological approaches, technological frameworks, and substantive insights garnered from the synthesized literature.

5.1. Overview of Collected Studies and Data

Information pertaining to the publication category of each study (such as Conference or Journal) was systematically gathered. The categorization based on the type of publication venue, along with the year-wise distribution of the relevant studies, is summarized in

Table 6. An overview of the studies publication type and year.

	2010	2011	2015	2017	2018	2019	2020	2021	2022	2023	# of Studies
Conference	1	1	1	2	1	3	1	1	0	2	13
Journal	0	0	0	0	0	0	0	2	2	1	5
Total	1	1	1	2	1	3	1	3	2	3	18

.

5.2. Generating Process Models from Natural Language Text Input (RQ1)

With research question RQ1, our aim is to elucidate the contemporary modeling technologies and methodologies that are applicable for modeling process models from natural language text input. Out of the 18 studies reviewed on the use of chatbots for process modeling, 15 focused on strategies for creating process models from natural language text input. These studies delineated three distinct methodologies to achieve this objective, as summarized in

Table 7. Distribution of the studies by technologies for generating process models from natural language text (RQ1).

Technologies	Studies	# of Studies
RQ1$A_1$: NLP	S1 [12], S2 [13], S3 [14], S5 [16], S6 [17], S8 [19], S9 [20], S10 [21], S11 [22], S12 [23], S13 [24], S14 [25], S15 [26], S17 [28]	14
RQ1$A_2$: Mapping Rules	S11 [22]	1
RQ1$A_3$: Sentence Level Analysis	S16 [27]	1

. The distribution of the used methods for RQ1 is presented in Figure 4.

Notably, NLP emerged as the predominant methodology, with 14 out of 15 studies leveraging NLP techniques to facilitate the conversion of natural language text input into structured process models. Additionally, Study S11 [22] mapped rules besides NLP, and Study S16 [27] used sentence-level analysis to accomplish this task.

We provide detailed summaries of the 15 studies in this domain, each elucidating distinct problems, solutions, and methodologies. Through a comprehensive analysis of these contributions, we aim to clarify the commonalities, differences, and emerging trends (cf. Figure 5).

In the realm of process modeling, extracting process models manually from textual data poses challenges in terms of time consumption and error susceptibility. However, ‘Extracting process models using NLP Technique’ [12] pioneers a solution by employing NLP techniques. Through syntactic and semantic analysis, Study S1 automates the extraction process, identifying process elements within text and converting them into structured models.

Another hurdle arises in the complex task of creating business process diagrams from textual descriptions. Study S2, ‘A Proposed Technique for process modeling Diagram using NLP’ [13], tackles this challenge by offering a methodological shift. Leveraging probabilistic latent semantic analysis, the approach developed by study S2 streamlines diagram creation by managing text complexity, effectively translating textual data into comprehensible business process diagrams.

Integrating sentiment analysis within business processes for understanding customer feedback presents yet another challenge. Study S3, ‘A Combined Model of NLP with process modeling for Sentiment Analysis’ [14], confronts this by merging NLP with process modeling. Analyzing textual data within the context of business processes enhances sentiment analysis, thereby improving the accuracy and relevance of insights.

Further streamlining the modeling process, study S5, ‘Assisted Declarative Process Creation from Natural Language Descriptions’ [16], introduces a tool to automate the creation of declarative process models from text. Utilizing NLP, this approach identifies and maps process constraints from textual descriptions, generating declarative process models efficiently.

By combining collaborative techniques with process mining, study S6, ‘A Case Study on Designing Business Processes Based on Collaborative and Mining Approaches’ [17], provides practical insights, demonstrated through a case study on the Ghana cocoa supply chain.

Easing the transition from textual requirements to formal process models, study S8, ‘Generating BPMN Diagram from Textual Requirements’ [19] introduces automation through NLP. By interpreting textual requirements and generating corresponding BPMN diagrams, this study facilitates a seamless visualization of business processes.

Ensuring accuracy in process element identification poses a challenge addressed by study S9, ‘A Visual Approach for Identification and Annotation of Business Process Elements in Process Descriptions’ [20]. Proposing a user-interactive method, study S9 enhances accuracy by employing visual tools for identifying and annotating process elements more precisely.

Efforts towards automation persist in study S10, ‘Extracting Declarative Process Models from Natural Language’ [21], where the manual extraction of declarative process models is transformed into an automated process using tailored NLP techniques. By identifying activities and constraints, this methodology generates declarative process models efficiently from textual data.

For identifying business process elements, study S11, ‘A Semi-automatic Approach to Identify Business Process Elements in Natural Language Texts’ [22], combines manual and automated approaches. By utilizing empirical mapping rules and NLP, this method semi-automatically identifies process elements from text, balancing efficiency with accuracy.

Study S12, ‘Supporting the Process of Learning and Teaching Process Models’ [23], introduces evaluation tools to aid process modelers creating accurate process models. By providing feedback on the correctness of process models, this approach enriches the learning experience, facilitating better comprehension and retention.

Addressing the challenge of process model discovery from unstructured text, study S13, ‘Automatic Process Model Discovery from Textual Methodologies’ [24], focuses on verb semantics. By leveraging NLP to mine process models, this study offers insights into structuring process elements from textual data.

Real-time process modeling during discussions is simplified by study S14, ‘Live Process Modeling with the BPMN Sketch Miner’ [25]. Through the development of a tool using NLP, study S14 enables users to generate and refine BPMN diagrams interactively, fostering collaboration and efficiency in process design.

Efforts to automate the generation process persist in study S15, ‘Process Model Generation from Natural Language Text’ [26]. By employing syntactic and semantic analysis, this methodology converts textual descriptions into process models, streamlining the modeling process and improving efficiency.

Proposing a conceptual solution, study S16, ‘A Concept for Generating process models from Natural Language Description’ [27], bridges the gap between textual descriptions and process models. By utilizing intermediate representations, this approach facilitates the conversion process, enhancing understanding and accuracy.

Innovation continues in study S17, ‘A Machine Translation Like Approach to Generate process model from Textual Description’ [28], where a machine translation-like approach is employed. By leveraging semantic transfer and syntactic manipulations, this study automates the conversion process, simplifying the transition from text to process models.

• Commonalities between the selected studies.

NLP techniques. Studies like ‘Extracting process models using NLP Technique’ (S1), ‘A Proposed Technique for process modeling Diagram using NLP’ (S2), and ‘Generating BPMN Diagram from Textual Requirements’ (S8) employ NLP to analyze and process text for creating process models.

Focus on automation. Studies such as ‘A Combined Model of NLP with process modeling for Sentiment Analysis’ (S3) and ‘Assisted Declarative Process Creation from Natural Language Descriptions’ (S5) aim to automate the conversion of text to process models, or vice versa.

Improving accuracy. Studies like ‘Extracting Declarative Process Models from Natural Language’ (S10), ‘A Semi-automatic Approach to Identify Business Process Elements in Natural Language Texts’ (S11), and ‘Process Model Generation from Natural Language Text’ (S15) focus on improving the accuracy and semantic consistency of the models or descriptions.

Evaluation metrics. Several studies, including ‘Supporting the Process of Learning and Teaching Process Models’ (S12), and ‘A Machine Translation Like Approach to Generate process model from Textual Description’ (S17), use various metrics to evaluate the accuracy and effectiveness of their methods.

• Differences between the selected studies.

Application focus. Studies like ‘A Combined Model of NLP with process modeling for Sentiment Analysis’ (S3) focus on specific applications, such as sentiment analysis.

Methodologies. Different studies use varying methodologies. For example, ‘A Proposed Technique for process modeling Diagram using NLP’ (S2) uses probabilistic latent semantic analysis, while ‘Automatic Process Model Discovery from Textual Methodologies’ (S13) focuses on verb semantics. ‘Live Process Modeling with the BPMN Sketch Miner’ (S14) emphasizes interactive tools, whereas ‘A Concept for Generating process models from Natural Language Description’ (S16) uses intermediate representations.

Type of process models. Some studies, such as ‘Generating BPMN Diagram from Textual Requirements’ (S8) and ‘Live Process Modeling with the BPMN Sketch Miner’ (S14), focus on BPMN diagrams, while others, like ‘Extracting Declarative Process Models from Natural Language’ (S10) and ‘Assisted Declarative Process Creation from Natural Language Descriptions’ (S5), work on declarative models.

User interaction. Studies like ‘A Visual Approach for Identification and Annotation of Business Process Elements in Process Descriptions’ (S9) and ‘Live Process Modeling with the BPMN Sketch Miner’ (S14) emphasize interactive tools for process modeling, whereas others, such as ‘Automatic Process Model Discovery from Textual Methodologies’ (S13), focus on automated methods.

Domain specificity.Some studies, like ‘A Case Study on Designing Business Processes Based on Collaborative and Mining Approaches’ (S6), apply their methods to specific domains (e.g., the Ghana cocoa supply chain), while others are more general.

5.3. Generating Natural Language Text from Process Models (RQ2)

With Research Question RQ2 we want to examine technologies and methodologies for generating natural language text from process models.

Table 8. Distribution of the studies for generating natural language text from process models (RQ2).

Technologies	Studies	# of Studies
-	-	0

shows the studies’ distribution. However, despite diligent inquiry, no studies specifically addressing this aspect were identified in the literature corpus reviewed. Nevertheless, it is pertinent to note the presence of a singular study, S18 [29], that encompassed both directions of inquiry—generating process models from natural language text input and, conversely, deriving natural language text output from process models. This study, though not directly aligning with the focal point of RQ2, warrants attention, as it offers insights into bidirectional transformation processes within the realm of process modeling and natural language interaction. Consequently, RQ2 remains unexplored in the extant literature corpus, with the sole study addressing bidirectional transformations to be analyzed comprehensively in the subsequent section pertaining to RQ3.

5.4. Generating Natural Language Text Input from Process Models and Backwards (RQ3)

Research Question RQ3 delves into studies that encompass bidirectional transformations, addressing both the conversion of natural language text to process models (RQ1) and the generation of natural language text output from process models (RQ2). This comprehensive inquiry seeks to explore the interplay between textual input and graphical representations within the context of process modeling. While RQ1 and RQ2 focus on unidirectional processes individually, RQ3 endeavors to examine the holistic integration of these processes, thereby elucidating the synergistic effects and challenges inherent in bidirectional transformation endeavors. As aforementioned, concerning RQ2, we have identified a study (S18) [29] by Mendling et al. that indeed addresses both directions of transformation—converting natural language text to process models and generating natural language text output from process models. Remarkably, similarly to the findings for RQ1, this study prominently employs NLP as its primary technology. During the conversion from Text to Model, the study uses natural language analysis involving tokenization, part-of-speech tagging, and semantic role labeling to parse textual descriptions. It then identifies process-related concepts and maps them into structured formats, like BPMN process models. For the conversion from Model to Text, study S18 details methods to analyze existing process models to extract their components and relationships. Using template-based or generative NLP techniques, it then generates textual descriptions while ensuring semantic consistency to accurately reflect the original models. The study also presents practical applications and case studies demonstrating the efficiency and scalability of these techniques. By addressing both directions of conversion, the study aims to bridge the gap between human-readable descriptions and machine-interpretable process models, facilitating better communication, documentation, and optimization of business processes. In

Table 9. Distribution of the studies for generating natural language text input from business process models and backwards (RQ3).

Technologies	Studies	# of Studies
NLP	S18 [29]	1

the distribution of the studies for generating natural language text input from business process models and backwards is presented.

5.5. How Are Chatbots Included in the Processes of RQ1–RQ3 (RQ4)?

Among the 18 studies gathered for analysis, a mere two integrated chatbots into their methodologies (cf.

Table 10. Distribution of the studies for using chatbots (RQ4).

Using Chatbots	Studies	# of Studies
√	S4 [15], S7 [18]	2

). Surprisingly, none of these studies effectively harnessed the potential of chatbots for generating process models from natural language text, nor for the conversion of process models into natural language text. Studies S4 [15] and S7 [18] were the only two studies we found with our search string including a chatbot. Study S4, ‘Comparing Generative Chatbots Based on Process Requirements: A Case Study’ [15], evaluates chatbots to determine their effectiveness in executing and managing predefined business processes. The chatbots are not used for process modeling or generating natural language text from process models itself, but rather for the execution and adherence to business process tasks. Study S4 assesses how well these chatbots can follow BPMN standards, manage task sequences, and handle decision gateways, providing insights into their practical applicability in supporting business process execution and management. Study S7, ‘Quantifying Chatbots’ Ability to Learn Business Processes’ [18], focuses on evaluating the capability of chatbots to learn and execute business processes. Study S7 develops a method to measure this proficiency using process mining metrics. By analyzing a large dataset of customer service interactions, the paper assesses how well chatbots can adhere to process models, execute tasks correctly, and improve over time. The evaluation provides insights into the effectiveness of chatbots in supporting and automating business processes but not for process modeling and generating natural language text from process models.

6. Discussion

This section discusses the need for a chatbot being able to generate process models in an interactive manner in collaboration with the user (i.e., the process modeler or stakeholders).

6.1. Need for an Interactive Chatbot Generating Process Models

The presented SLR has shed light on the current state of research in the domain of using interactive chatbots for generating process models. Through the comprehensive analysis of scholarly articles, it becomes evident that there is a gap in the existing body of knowledge. No study has been found that perfectly aligns with the specific requirements for an interactive chatbot capable of generating and modifying process models to meet user needs.

The prevailing research trends primarily focus on distinct facets related to process modeling and chatbot technology. These trends include, but are not limited to, endeavors aiming at extracting information from textual inputs to generate process models or generating natural language text from process models. Nevertheless, there is a notable gap in the literature, as there is no single chatbot demonstrating both. The absence of such a comprehensive solution raises fundamental questions about the existing state of technology in this field. Investigating communication technologies in specialized settings, such as those highlighted in the study ‘Development of underground communication system for data transmission using Wi-Fi Direct and power line communication’ [30], could offer valuable insights for deploying advanced technologies like interactive chatbots. Understanding how hybrid communication systems perform in challenging environments can provide a broader context for enhancing technological applications in industrial or complex information systems. In essence, the SLR underscores the pressing need for a comprehensive solution—an interactive chatbot that possesses the capability to generate process models from natural textual input and adapt them in response to user requests but also for the opposite way. This unmet need represents a fertile ground for future research and development, opening up exciting opportunities to bridge the gap between chatbot technology and process modeling, thereby revolutionizing the way organizations handle their process management needs. Future research endeavors in this direction will undoubtedly contribute significantly to the fields of AI, LLM, NLP, and Business Process Management.

6.2. Experiment

We conducted a small-scale proof-of-concept experiment with two independent groups of students (a group of Bachelor’s students and a group of Leader students in the Software Engineering program) to explore the potential of using chatbots and LLMs for generating process models from natural language text. The goal was to determine whether, in principle, it is feasible to produce a representation that could be visualized, with common tools and then iteratively be refined through interactions with the chatbot.

6.2.1. Goal of the Experiment

The primary goal of the experiment was to evaluate whether current chatbot technology could be leveraged to generate process models from natural language input and to support iterative refinement of these models through further interactions with the chatbot. We aimed to determine the feasibility of this approach, rather than focus on the quality or accuracy of the resulting models, which would require more extensive research in future work.

6.2.2. Study Design

Two independent groups, each consisting of two students, worked on this project. One group focused on developing a prototype capable of generating an initial process model from natural language input with the help of the API from OpenAI https://platform.openai.com/api-keys (accessed on 1 September 2024), while the second group concentrated on refining and adapting an existing model based on user-provided textual instructions. Both groups explored the chatbot’s ability to handle tasks such as generating an initial process model, modifying an existing process model based on textual descriptions, and answering questions about the model. The models were then visualized using BPMN.iO https://bpmn.io (accessed on 1 September 2024).

6.2.3. Study Results

The experiment yielded promising results, demonstrating that chatbots can indeed generate basic process models from natural language text. The prototypes were able to create initial models, modify them through textual descriptions, and even handle simple user queries regarding the models. These results show that, in principle, current chatbot technology has the potential to support this type of interaction, sometimes with surprisingly good outcomes. However, the study also revealed several limitations. More complex scenarios, such as multi-party processes involving multiple pools or message flows, were not adequately handled by the prototypes. Additionally, while the basic functionality works, further research is needed to assess the quality of the models and improve the handling of more sophisticated process features. To illustrate the progress made and provide a clearer picture of what is currently feasible with chatbot technology, Figure 6 and Figure 7 present screenshots of the prototype.

What works well: Chatbots can generate and modify simple process models, support interactive refinement, and answer questions about the model. This shows that the technology has the potential to assist in process model generation and iteration, at least for basic use cases. Hence, further research on the use of chatbots generating process models can be justified.

What needs further research: More complex process modeling tasks, such as handling multi-party processes, message flows, and multi-pool models, still require further development. Additionally, there is a need for defining clear evaluation criteria regarding the quality of the generated models. Note that this goes beyond simple accuracy and touches on aspects like model comprehensibility, correctness, and suitability for practical use. To ensure the accuracy of chatbot-generated models, it is crucial to implement rigorous validation processes, such as comparing chatbot outputs with expert-generated models and using benchmarking against established standards. Additionally, regular audits and incorporating feedback loops can help in refining the models over time. Concerns regarding data security and privacy are critical, requiring robust encryption methods and adherence to data protection regulations to safeguard sensitive information. Additionally, implementing strict access controls, conducting regular security audits, and ensuring anonymization of data where possible can further enhance protection. It is also crucial to integrate secure data handling practices and maintain transparency with users about data usage policies to build trust and ensure compliance with privacy laws. We believe that these initial results confirm the potential of chatbot-based process model generation but also highlight the need for further research to evaluate the quality of the models and to develop methodologies to handle more complex scenarios. In summary, while our proof-of-concept experiment shows that chatbots can indeed generate process models from natural language text and interactively refine them, further research is needed to fully unlock their potential and ensure the quality and accuracy of the resulting models.

6.3. Directions for Future Research

As the field of BPM evolves, the integration of interactive chatbots represents a promising avenue for advancing all stages of the BPM lifecycle (cf. Figure 8). Through brainstorming with process modeling and AI experts, we discussed the potential impact of integrating interactive chatbots into each stage of the BPM lifecycle and outline future research directions in this area.

Process identification. Future research could delve into how interactive chatbots can aid in identifying relevant business processes by actively engaging users to gather process-related information from diverse sources. Investigating the effectiveness of NLP and ML algorithms in extracting process details from user conversations and textual data offers promising perspectives for enhancing process identification methods.

Process discovery. In the realm of process discovery, leveraging interactive chatbots to facilitate collaborative discussions and brainstorming sessions among stakeholders could improve the way processes are uncovered. Exploring techniques for integrating user-generated content from chatbot interactions with process mining algorithms offers opportunities for more dynamic and context-aware process discovery approaches.

Process analysis. Future research may deal with interactive chatbots that provide on-demand process analysis and insights to users. Investigating how chatbots can identify process bottlenecks, analyze performance metrics, and recommend improvements based on user feedback holds potential for enhancing decision making and optimization efforts.

Process redesign. Exploring how interactive chatbots can facilitate collaborative process redesign efforts by soliciting feedback from users and guiding them through the redesign process is a promising research direction as well. Investigating the integration of chatbots with design thinking methodologies could streamline process redesign efforts and foster stakeholder engagement.

Process implementation. Leveraging interactive chatbots to automate the deployment and configuration of process models based on user specifications and requirements presents another intriguing avenue for future research. Exploring techniques for integrating chatbots with workflow automation platforms and business process management systems could streamline implementation processes and enhance user adoption.

Process monitoring and control. Developing interactive chatbots that provide real-time monitoring and alerts for process deviations and exceptions is a compelling area for future research. Investigating how chatbots can facilitate proactive process management by notifying users of potential issues and guiding corrective actions (i.e., prescriptive process monitoring [31]) could enhance process performance and compliance.

Process optimization. Future research directions may involve leveraging interactive chatbots to support continuous process improvement efforts by actively engaging users in optimization activities. Exploring how chatbots can analyze user feedback, identify optimization opportunities, and recommend personalized enhancements could enhance the effectiveness of optimization efforts.

Process automation. Exploring the role of interactive chatbots as intelligent process agents orchestrating automated process executions based on user input and preferences presents exciting research opportunities. Investigating the integration of chatbots with robotic process automation (RPA) tools and autonomous execution platforms could lead to more adaptive and context-aware automation solutions.

7. Threads to Validity

To assess the veracity of the SLR and to discern potential vulnerabilities therein, we adhere to the classification schema for secondary studies established by Ampatzoglou et al. [32]. This schema encompasses three primary facets: study selection validity, data validity, and research validity.

Study selection validity is primarily concerned with validating the procedure employed for the acquisition and filtration of pertinent scholarly works [32].

The selection of digital libraries and the formulation of search strings wield considerable influence over the resultant findings and, thus, necessitate meticulous deliberation. In the methodology, we have opted for prominent libraries such as ScienceDirect, IEEE Explore, Web of Science, Springer Link, and ACM Digital Library. However, it is imperative to recognize that this might not universally apply to all relevant works. Despite our painstaking efforts to devise a comprehensive search query, we cannot guarantee its comprehensive inclusivity in capturing all pertinent research. In response to this potential limitation, we have introduced a snowballing mechanism [9] to encompass research outputs that may have eluded detection through the initial query.

Another facet pertaining to validity emanates from our pre-established set of selection criteria. Nonetheless, we remain sanguine regarding the pertinence of our findings, which center predominantly on the methodology phase, and anticipate their capacity to offer a nuanced perspective and meaningful contribution to the extant body of literature.

The concept of data validity pertains to the validation of data extraction and the subsequent analysis of secondary studies [32]. Within this category, various threats can manifest, including biases in data collection and analysis. It is noteworthy that different researchers may interpret a study in distinct ways, particularly when it comes to categorizing the extracted data. In order to address these potential biases comprehensively, during the data extraction phase, we checked the studies online in detail to facilitate the sharing and deliberation of extracted data among the researchers involved https://docs.google.com/spreadsheets/d/1ULUpikNz-VY2ezi6ybKViT-lmWoZ38bHgImNx2jNPIs/edit?usp=sharing (accessed on 1 September 2024). Additionally, we conducted a cross-check process involving continuous reviews of the data by researchers who were not directly engaged in the initial data extraction. In cases where conflicts or discrepancies arose during this cross-check phase, they were resolved through collaborative discussions.

The concept of research validity is concerned with validating all aspects of the secondary research methodology, encompassing the coverage of research questions and the repeatability of the survey [32]. In an effort to mitigate the threat to research coverage, we took several measures. In Section 3.1, we delineated four primary research questions, with a specific focus on existing technologies. The analysis predominantly focused on an approach, with a primary emphasis on the technology. To assure the repeatability of our work, as outlined in Section 2 and Section 3, we provided a detailed description of the search protocol, which delineates the systematic process utilized in our research.

8. Conclusions

This paper presents the findings of a systematic literature review focusing on the topic of interactive chatbots in the context of generating process models. The SLR was conducted in accordance with Kitchenham’s guidelines [8]. To enhance its comprehensiveness, both backward and forward snowballing [9,10] were employed. A detailed description of the methodology used in this study has been provided to ensure the reproducibility of the research process. After conducting search and filter procedures, a total of 18 studies were selected and subjected to analysis. The paper commences by establishing the significance of interactive chatbots in the generation of process models. Subsequently, the methodology employed is discussed, followed by the organization of the papers sections to align with the systematic protocol, including planning, conducting and reporting. In addressing the research questions, it becomes evident that interactive chatbots involved in process model generation constitute a prominently discussed topic. They are primarily utilized for extracting knowledge from textual inputs and translating it into a process model. Furthermore, these chatbots often serve as tools for responding to inquiries related to pre-modeled process models. Finally, the paper outlines several promising research directions for future investigations in the realm of interactive chatbots engaged in the generation of process models. These directions aim to guide and inspire further scholarly endeavors in this evolving field.