This paper introduces AIRSim (AI Responses Simulator), an innovative AI tool designed to support students in practicing their questionnaire analysis skills within the café and restaurant discipline. Utilizing artificial intelligence (AI), AIRSim generates hypothetical feedback data to facilitate student learning. Through a series of 16 experiments, we evaluated AIRSim's capability in simulating participant responses to user-uploaded questionnaires. Our findings demonstrated a notable degree of diversity in the generated results, as indicated by the Entropy Index, across various perspectives and participant-question combinations. To the best of our knowledge, there exists a lack of relevant studies exploring this specific application of AI in the context of student learning within the café and restaurant discipline. By introducing the AIRSim tool, educators can efficiently enhance their students' analytical abilities and responsiveness to customer needs. This practical contribution addresses the pressing need for effective training methods in the hospitality sector while also capitalizing on the transformative potential of Generative AI technologies, such as ChatGPT. Overall, this study provides valuable insights into AI-driven student learning and identifies areas for future research.
Research Article
Introducing AIRSim: An Innovative AI-Driven Feedback Generation Tool for Supporting Student Learning
https://doi.org/10.21203/rs.3.rs-4365670/v1
This work is licensed under a CC BY 4.0 License
Version 1
posted
You are reading this latest preprint version
Generative AI
Café and restaurant
Student learning
Simulation
ChatGPT
Questionnaires are widely used in café and restaurant businesses to gather valuable customer feedback. Through the analysis of such feedback, businesses can derive significant insights that greatly contribute to the enhancement of customer satisfaction, the improvement of service quality and the cultivation of a customer-centric environment.
To fully capitalize on the benefits of customer feedback, it is crucial for businesses to provide relevant training to their staff. Previous studies have indicated that staff training is indispensable in the café and restaurant industry, offering a range of benefits from improved service quality and customer satisfaction [1–4] to increased efficiency [5, 6] and innovation [7, 8].
Specifically, in today's rapidly changing world, people's thoughts and preferences evolve swiftly, making the ability to adapt quickly to these changes crucial. Therefore, related training should focus on equipping staff with the skills necessary to adeptly navigate the analysis process, rather than merely concentrating on end results. This training focus holds significant importance for several reasons. Firstly, it ensures long-term sustainability by enabling staff to consistently derive insights from various datasets over time. Secondly, it encourages adaptability, allowing them to adjust their approach to different types of feedback data. Thirdly, it enhances comprehension of analysis principles, thereby improving decision-making and insight provision. Moreover, it fosters a culture of continual improvement, empowering staff to refine their skills and explore new analysis methods.
Higher education should equip relevant students with the skills necessary to adeptly navigate the analysis process, recognizing that these students are the future staff of businesses. Such skills not only ensure long-term sustainability but also foster adaptability, enhance decision-making, and cultivate a culture of continual improvement. By prioritizing training focused on analysis proficiency, higher education institutions prepare students to derive valuable insights from diverse datasets, adapt their approach to evolving circumstances, and contribute effectively to their future roles in businesses, thus ensuring their preparedness for professional success and the advancement of their respective industries.
In order to prioritize educational endeavors aimed at equipping students with proficiency in the analysis process, educators should expose students to abundant and varied feedback data. More data provides a broader spectrum of insights and scenarios for students to analyze, which in turn fosters a deeper understanding of analysis principles and improves decision-making abilities. When students are exposed to diverse feedback data, they encounter a wider range of customer experiences, preferences and challenges. This exposure enables them to develop robust analytical skills that can be applied across different contexts and situations. Additionally, working with abundant data allows students to explore various analysis techniques and methodologies, enhancing their adaptability and problem-solving capabilities. Ultimately, the more data students have access to during their training, the better prepared they will be to effectively analyze feedback and drive meaningful improvements in customer satisfaction and service quality.
Nonetheless, a significant challenge in educational practice is obtaining a sufficiently diverse range of feedback data. Obtaining diverse feedback data in higher education settings faces challenges on multiple fronts. Firstly, without industrial partners to share data, institutions lack access to real-world feedback that reflects industry diversity. Secondly, even with industrial partnerships, reluctance from partners to share data can impede efforts to gather diverse feedback. Thirdly, collecting primary data from students and staff can be prohibitively time-consuming and costly, hindering the acquisition of a broad range of perspectives. These challenges underscore the difficulty in obtaining sufficient and varied feedback data for effective analysis in café and restaurant businesses education.
To address the challenge of obtaining ample and diverse feedback data to support relevant education purposes, leveraging hypothetical feedback data can be a promising solution. Hypothetical data involves creating simulated feedback based on possible customer responses to various service aspects. This approach allows students to practice analyzing feedback and develop problem-solving skills without waiting for a large volume of real customer data. Additionally, it prepares the students to efficiently handle real feedback when it becomes available, ensuring a quicker and more effective response to customer needs.
This study presents an innovative AI tool, AIRSim (AI Responses Simulator). The tool leverages artificial intelligence (AI) to generate hypothetical feedback data for students to practice their analysis skills. To the best of our knowledge, there exists a lack of relevant studies exploring this specific application of AI in the context of student learning within the café and restaurant studies. The absence of prior research in this specific domain highlights a significant research gap. By identifying and addressing this gap, the present study contributes to the advancement of knowledge in the field of café and restaurant education methodologies, offering a novel perspective on utilizing AI to enhance student learning practices.
In practical terms, this tool offers multifaceted advantages. Primarily, it furnishes educators with a broader dataset to train their students, encompassing a diverse range of customer interactions and scenarios. This equips students with the requisite skills to adeptly navigate various situations. Secondly, it mitigates the reliance on real-time customer feedback, which can be both constrained and time-intensive to gather. Theoretically, this study contributes to the domain of AI applications in business training, exemplifying how AI can be ingeniously harnessed to augment learning outcomes. Furthermore, it advances the understanding of scenario-based training methodologies by introducing a novel mechanism for generating training data. In summation, this innovative approach holds promise for enhancing the effectiveness and efficiency of student learning in café and restaurant establishments.
The remaining parts of this paper are organized as follows. The literature review section gives an overview of three important areas: Generative AI and text generation, question-answering with Generative AI and the advancements shown by ChatGPT. This review also sets the foundation for suggesting a tool that combines these key elements. Moving forward, the subsequent sections will delve into the design, functionality, experiments and analysis of results to evaluate the performance of our AI tool. Through these endeavors, we aim to contribute both theoretically and practically to the field, while also acknowledging the limitations encountered and suggesting potential avenues for future research and development.
This section provides an overview of three key areas: Generative AI and text generation, question-answering with Generative AI and the advancements represented by ChatGPT. The aim of this review is to set the stage for proposing a tool built upon these key areas. This tool aims to integrate the robust capabilities of Generative AI, particularly ChatGPT, to develop a versatile system for generating responses to open-ended questions within unstructured questionnaires. By leveraging AI text generation, question-answering features and role-play functionalities, the proposed tool seeks to offer practical applications in diverse settings, enhancing user experiences and driving better outcomes across various sectors.
2.1 AI text generation
The remarkable capabilities of Generative AI in text generation span across diverse domains. A previous study [9] underscores the practical implications of Generative AI in streamlining professional tasks. Through autonomous drafting of emails and other correspondence, AI models significantly reduce the time required for such activities across various industries. This efficiency not only enhances productivity but also allows professionals to allocate their time more strategically. In customer service, the impact of Generative AI is evident in its ability to deliver realistic and contextually appropriate responses [10]. By simulating human-like interactions, AI-powered chatbots elevate customer experiences, fostering engagement and satisfaction. In marketing, Generative AI generates content personalized to individual preferences and behaviors. As per Kaplan and Haenlein, this personalized approach fosters a deeper connection with consumers, leading to more effective communication and ultimately driving better business outcomes. Moreover, in the legal sector, Gao et al. [12] discuss how Generative AI facilitates the creation of legal documents, contracts and briefs. Its capability to analyze legal texts for relevant information contributes to time savings and reduces the burden of manual labor, thereby increasing efficiency within legal workflows. Furthermore, Liu et al. [13] highlight the role of Generative AI in the development of summarization tools. By condensing extensive articles, reports and papers into concise summaries, these tools aid in efficient information digestion for users across various fields. This not only saves time but also enables users to access key insights more rapidly, thus enhancing decision-making processes.
Expanding upon the remarkable capabilities of generative AI in text generation, this study focuses on Generative AI's application for generating responses to complete open-ended questions within unstructured questionnaires. Leveraging the question-answering capabilities inherent in Generative AI, this methodology aims to further explore its potential in practical applications.
2.2 AI question-answering
The applications of Generative AI in question-answering have extensively been documented across various domains.
In educational settings, Generative AI aids students by offering study-related assistance, explaining complex subjects and providing feedback on exercises, thereby enhancing learning outcomes [14]. Similarly, in healthcare, Generative AI provides quick responses to medical inquiries, suggests diagnoses and offers guidance on treatment options, drawing from extensive medical literature to make healthcare advice accessible and timely [15].
Customer support services have also benefited significantly from Generative AI, as it empowers chatbots and virtual assistants to promptly respond to inquiries, thereby improving operational efficiency and user experience [10]. Moreover, in legal contexts, Generative AI facilitates legal research and compliance activities by addressing queries on legal interpretations and case law, aiding legal professionals in navigating complex documents [16]. Furthermore, Generative AI drives technical support systems by providing step-by-step troubleshooting guides and solutions to common issues, thus enhancing user support without human intervention [17]. In the tourism industry, AI-driven question-answering systems offer personalized recommendations, travel information and assistance with travel logistics, enriching the travel planning process [18, 19]. Additionally, Generative AI extends its utility to financial and risk management applications [20, 21].
In summary, Generative AI plays a pivotal role in question-answering across various sectors, facilitating efficient responses, enhancing user experiences and democratizing access to information.
2.3 ChatGPT and role play
In recent years, advancements in artificial intelligence have led to the development of ChatGPT, a language model with strikingly human-like abilities. Various studies have explored its capabilities in mimicking human conversation convincingly, demonstrating exceptional intelligence and showing nuanced emotional responses akin to humans. Additionally, research has delved into its potential applications across different fields, suggesting a wide range of possibilities for its use.
Recent studies have highlighted the impressive human-like capabilities of ChatGPT across various cognitive areas. Biever [22] notes ChatGPT's achievement in breaking the Turing test, a milestone indicating its capacity to mimic human conversation convincingly. A study [23] further demonstrates ChatGPT's exceptional performance on intelligence tests, surpassing the IQ scores of individuals with advanced degrees or even university professors. Additionally, Ji et al. [24] highlights ChatGPT's consistent ranking preferences, aligning closely with human behavior. Moreover, Orrù et al. [25] showcases ChatGPT's capability in out-of-the-box thinking, particularly in solving verbal insight problems, showcasing its creativity akin to human cognition. Leong and Sung [26] suggests that ChatGPT exhibits preferences akin to humans, favoring input messages presented in positive tones over negative ones. Furthermore, a finding [27] on ChatGPT-4 bots reveal distinct response patterns influenced by primed emotions, indicating nuanced decision-making reminiscent of human emotional processing. Collectively, these studies emphasize ChatGPT's remarkable human-like abilities, ranging from conversational prowess to cognitive reasoning and emotional responsiveness.
In addition, recent research highlights ChatGPT's potential in various role-playing scenarios. In language education, preliminary findings [28] suggest that ChatGPT serves as a beneficial tool for acquiring formal register knowledge, although its necessity for informal writing remains questionable. Additionally, in agricultural contexts, ChatGPT has been successfully applied as a chatbot assistant, outperforming traditional methods in providing quality responses to farmers' inquiries [29]. Moreover, investigations into ChatGPT's accuracy in offering nutritional information for dietary planning reveal promising results, positioning it as a reliable AI dietician [30]. Furthermore, collaborative approaches involving both AI and human experts have shown improvements in content appropriateness and readability, without compromising clinical detail [31]. In customer service, ChatGPT demonstrates effectiveness in generating management responses to customer reviews on platforms like Tripadvisor, meeting the requirements of efficient and satisfactory customer interactions [32]. Stampfl, Ivkić and Geyer [33] explores its use in a simulation game, revealing enhanced teaching quality and promoting active learning. Chen et al. [34] introduces RoleInteract, a tool to assess the sociality of role-playing conversational agents, benefiting both individual and group interactions. Additionally, Maurya [35] introduces ChatGPT as a simulated client for practicing counseling skills, with results indicating its authenticity, consistency, emotional expression, cultural sensitivity, empathy and self-awareness, enhancing the learning experience.
2.4 Summary
Across diverse domains such as creative writing, customer service, legal documentation and education, Generative AI has revolutionized text generation, streamlining professional tasks and enhancing user experiences. Similarly, its role in question-answering has facilitated efficient responses and democratized access to information across various sectors.
The remarkable capabilities of ChatGPT, highlighted in recent studies, showcase its human-like conversational abilities, cognitive reasoning and role-play functionalities. These advancements underscore the potential for integrating ChatGPT into practical applications, offering personalized, contextually appropriate responses to open-ended questions within unstructured questionnaires.
Building upon these key elements, we propose an AI tool, AIRSim (AI Responses Simulator), which seeks to harness the strengths of Generative AI and ChatGPT to create a versatile system for generating varied responses. By leveraging AI text generation, question-answering features and role-play capabilities, the tool aims to enhance student learning experience in terms of practicing questionnaire analysis.
The tool design and functionality are discussed in next section.
The AIRSim, built upon the capabilities of ChatGPT, serves as the foundational infrastructure for its development and deployment, integrating advanced AI text generation and question-answering functionalities. This amalgamation enables the simulation of participant responses within user-provided questionnaires. The AIRSim is available to subscribers of ChatGPT plus through the OpenAI’s GPTs platform which can be accessed through this link: https://chat.openai.com/g/g-vFHpcJuiX-airsim.
This section outlines the basic functions of the tool, explaining how it interacts with users and the processes involved in generating simulated responses.
Interaction with users
The tool engages users through a structured interaction sequence aimed at eliciting necessary inputs and specifications for generating simulated participant responses. As per Fig. 1, the interaction protocol encompasses the following steps:
Questionnaire Upload: Users are prompted to upload the questionnaire they intend to administer to simulated participants. This questionnaire serves as the template for generating simulated responses.
Participant Quantity Specification: Users specify the desired number of simulated participants they wish to generate for questionnaire completion. This input parameter governs the scale of simulated data generation.
Participant Profile Creation: The tool dynamically creates simulated participants based on user-specified criteria. Each participant is endowed with a unique combination of demographic characteristics, including nationality, gender and age. These attributes imbue each participant with distinct identities, facilitating diverse responses reflective of real-world variability.
Questionnaire Response Generation: Leveraging the user-uploaded questionnaire and the predefined participant profiles, the tool orchestrates the generation of simulated responses. Each participant's responses are tailored to their individual characteristics, ensuring nuanced and contextually relevant answers. Consequently, disparate participants offer diverse perspectives and insights, enriching the dataset with multifaceted responses.
Data Compilation and Export: Upon generating simulated responses, the tool collates the participant data into a structured format. A new Excel file is generated, featuring designated columns for participant identification, demographic attributes (i.e., nationality, gender, age) and individual question responses. Each row corresponds to a distinct participant, encapsulating their comprehensive questionnaire responses.
In order to evaluate the performance of our AI tool, we conducted a series of 16 experiments for the purpose. The primary objective of these experiments was to assess the tool's capability in generating simulated participants and diverse responses to user-uploaded questionnaires. Each experiment varied in the number of questions within the questionnaires and the quantity of simulated participants generated.
- Questionnaire Characteristics:
The questionnaires utilized in the experiments were designed as typical survey instruments. They were structured to contain 5, 10, 15 and 20 questions, enabling us to evaluate the tool's performance across different questionnaire lengths. Table 1 lists the 20 experiment questions.
- Participant Characteristics:
For simplicity and consistency across experiments, all questionnaires included three characteristic attributes: Nationality, Gender and Age. These attributes were chosen due to their fundamental relevance in demographic surveys and to ensure a baseline for comparison across experiments.
Experiment Parameters
Each experiment involved a combination of questionnaire length and participant quantity, as detailed below:
Questionnaire Lengths: The questionnaires were uploaded with 5, 10, 15 and 20 questions.
Participant Quantities: The tool was tasked with generating simulated responses from 10, 50, 100 and 500 participants for each questionnaire. These varying quantities allowed for the evaluation of scalability and performance under different workload conditions.
Evaluation Methodology
To assess the diversity of the generated results produced by the tool in each experiment, we employed entropy computation. Entropy is a measure widely used in information theory to quantify uncertainty or randomness in a dataset. Higher entropy values indicate a more diverse dataset, where outcomes are spread out across a wider range of possibilities. Conversely, lower entropy values suggest a more uniform distribution, with outcomes clustered around a few dominant characteristics or responses.
The formula for computing entropy, commonly denoted as H, is as follows:
Where:
- H(X) represents the entropy of the dataset X.
- P(Xi) denotes the probability of occurrence of the ith outcome in the dataset.
- n signifies the total number of distinct outcomes in the dataset.
In the context of our experiments, entropy computation enabled us to evaluate the spread and diversity of responses across different participant characteristics and questionnaire attributes.
- Experimental Procedure:
- Setting Experiment Parameters: Prior to each experiment, we defined the parameters including the length of the questionnaire and the quantity of simulated participants.
- Questionnaire Upload: The designed questionnaires were uploaded to the tool, specifying the desired participant characteristics and response quantities.
- Simulated Participant Generation: The tool generated simulated responses from the specified number of participants, considering the defined characteristics and ensuring diversity in the generated dataset.
- Entropy Computation: Post-generation of responses, we computed the entropy of the dataset to quantify the diversity of responses across the specified characteristics.
- Data Analysis: The entropy values obtained from each experiment were analyzed to evaluate the tool's performance in generating diverse and representative responses.
The 16 experiments were conducted on 22 March 2024. Each experimental iteration yielded results were categorized into three distinct perspectives: i) demographic data, ii) questionnaire responses and iii) observations. The demographic data encompasses information such as nationality, gender and age, automatically generated by AIRSim for each entry. Questionnaire responses pertain to the answers provided by AIRSim to individual questions posed within the questionnaire. The observation constitutes the entirety of the amalgamation between demographic data and questionnaire responses.
-
Entropy index
To quantify the diversity within each perspective, the Entropy Index was introduced. This index was derived by dividing the calculated entropy of the respective perspective by its maximum possible value. The resulting index ranges from 0 to 1, where 0 signifies minimal diversity and 1 denotes maximal diversity. Figure 2 through 4 elucidate the Entropy Index across the 16 experiments, each bar represents an experiment result of a participant-question combination (P and Q). They offer insights into the diversity levels across each perspective, namely demographic data, questionnaire responses and observations.
The findings depicted in the diagrams reveal a noteworthy degree of diversity within the generated results of AIRSim, with Entropy Indices either equal to 1 or in close proximity across each perspective and participant-question combination (P and Q).
-
Average Entropy index
In addition to the aforementioned assessments pertaining to the amalgamation of demographic data and questionnaire responses, further analysis was conducted by calculating the entropy index for each individual item, encompassing attributes such as nationality, age, gender, as well as responses to specific questions (e.g., Q1, Q2 and subsequent queries). Subsequently, the average entropy index was computed for demographic attributes, questionnaire responses and the overall observation.
The rationale for this additional analysis is that when questions are grouped into categories, there is a risk of losing granularity in the analysis. Different questions within a category may exhibit varying levels of diversity. Aggregating them together obscures these differences and may result in oversimplification of the diversity landscape.
Conducting further analysis by calculating entropy at the individual item level followed by obtaining the average entropy index, helps to mitigate the issue of loss of granularity by providing detailed insights into diversity across specific attributes. This is because by calculating the entropy index for each individual item, granularity in the data is retained. This allows for detailed insights into the diversity present within each specific attribute, whether it pertains to nationality, age, gender, or responses to specific questions. Retaining this granularity ensures that unique variations and patterns within each attribute are evaluated.
Figure 5 through 7 visually depict the average entropy indices across the 16 experiments. This augmented analytical approach, encompassing the calculation of entropy indices for individual columns and the subsequent derivation of average entropy indices for demographic attributes, questionnaire responses and the holistic observation, engenders a more nuanced comprehension of diversity within distinct dimensions of the dataset. Such methodological refinement bolsters the thoroughness of our analytical framework and enriches the insights gleaned from the evaluation of the results generated by AIRSim.
As evidenced by the diagrams, the observed entropy indices were generally lower compared to those depicted in Fig. 2 through 4. This discrepancy can be attributed to the constrained variability inherent in certain attributes, for example, the item gender exhibits limited diversity. Consequently, when entropy indices are computed on a per-item basis, attributes with restricted variability exert a discernible influence on the overall diversity of results. Nevertheless, these findings illuminate a noteworthy trend: experiments involving a minimum number of participants tend to yield higher entropy indices (i.e., P = 10), whereas experiments with a larger participant pool tend to yield lower entropy indices (i.e., P = 500). Analogous trends were consistently observed across all 16 experiments.
In the café and restaurant industry, customer feedback is crucial for improving service quality and satisfaction. Teaching students focusing on analysis skills proves indispensable in navigating the evolving landscape, ensuring sustained insights, adaptability, and improved decision-making. However, educators often face challenges in obtaining diverse feedback data. To address this, the utilization of hypothetical feedback data emerges as a promising solution, enabling students to hone their analytical abilities efficiently. By incorporating hypothetical data into training regimes, educators can better prepare their students for real-world scenarios, fostering quicker and more effective responses to customer needs. Thus, integrating hypothetical feedback data into training programs stands as a strategic approach to overcoming limitations in data availability, ultimately contributing to the continual improvement of customer-centric environments in the hospitality sector.
In this study, we introduced AIRSim, a novel tool leveraging artificial intelligence (AI) for generating hypothetical feedback data to aid in student learning within café and restaurant businesses. In recent years, the emergence of Generative AI has transformed numerous fields, such as creative writing, customer service, legal documentation and educational assistance. Its impressive text generation and question-answering capabilities have streamlined various professional tasks and improved user experiences across diverse sectors. Notably, ChatGPT, an advanced language model, exemplifies Generative AI, demonstrating human-like conversational skills and cognitive reasoning.
Through a series of 16 experiments, we evaluated AIRSim's capability in simulating participant responses to user-uploaded questionnaires. Our findings demonstrated a notable degree of diversity in the generated results, as indicated by the Entropy Index, across various perspectives and participant-question combinations.
Furthermore, one significant observation from our analysis was the inverse relationship between the number of participants and the entropy indices. Specifically, experiments with fewer participants tended to yield higher entropy indices, while experiments with a larger participant pool resulted in lower entropy indices. This trend underscores the importance of considering sample size when assessing the diversity and representativeness of simulated feedback data.
6.4 Theoretical and Practical Contributions
This work contributes theoretically by proposing a novel approach to addressing challenges in obtaining diverse feedback data in the café and restaurant industry. By introducing the concept of utilizing hypothetical feedback data for student learning, the study highlights a strategic method for enhancing analytical skills among students. The incorporation of hypothetical data into learning practice not only offers a practical solution to data availability limitations but also underscores the importance of adaptability and improved decision-making in navigating the dynamic landscape of customer feedback. Moreover, the study extends theoretical insights by leveraging advancements in Generative AI to develop a tool capable of generating realistic hypothetical feedback data, thus bridging the gap between theoretical concepts and practical implementation in student learning.
In practical terms, this work offers a tangible solution for educators in the café and restaurant discipline to improve service quality and customer satisfaction through student practice. By introducing the AIRSim tool, which utilizes artificial intelligence (AI) for generating hypothetical feedback data, educators can efficiently enhance their students' analytical abilities and responsiveness to customer needs. This practical contribution addresses the pressing need for effective training methods in the hospitality sector while also capitalizing on the transformative potential of Generative AI technologies, such as ChatGPT. By implementing AIRSim into their training regimes, students can expect quicker adaptation to real-world scenarios, leading to enhanced customer experiences and sustained competitiveness in the industry.
6.5 Limitations
Despite its theoretical and practical contributions, this work is not without limitations. One potential limitation lies in the reliance on hypothetical feedback data generated by AI. While these data may simulate real-world scenarios, they may not fully capture the nuances and variability of actual customer feedback. There is a risk that students trained solely on hypothetical data may encounter challenges in applying their skills to diverse and unpredictable situations encountered in real-life interactions with customers.
To address the limitation, educators could supplement AI-generated data with real customer feedback whenever feasible. This approach would provide students with a more holistic learning experience, allowing them to engage with authentic customer interactions and gain insights into real-world scenarios. Moreover, the learning approach should focus on equipping students with the skills necessary to adeptly navigate the analysis process, rather than merely concentrating on end results. By emphasizing the analytical process, students can develop critical thinking skills and learn how to interpret and analyze feedback effectively, regardless of whether it originates from AI-generated or real customer data. This approach not only enhances the students’ ability to adapt to diverse scenarios but also fosters a deeper understanding of customer needs and preferences, ultimately leading to improved service quality and customer satisfaction in the café and restaurant industry.
Furthermore, while Generative AI technologies such as ChatGPT offer remarkable capabilities in text generation and question-answering, they are not immune to biases inherent in the training data. There is a risk of perpetuating biases or generating inappropriate responses, which could undermine the effectiveness and credibility of the training provided by the AIRSim tool. On this regard, educators should systematically review and check the feedback generated by the AIRSim tool.
6.6 Potential Future Works
Building upon the foundation laid by this study, future research could explore additional applications of AI-generated feedback data in the café and restaurant education. One avenue for further investigation could involve refining the AI algorithms used to generate hypothetical feedback data, with a focus on enhancing the realism and diversity of the generated responses. This could involve incorporating natural language processing techniques to better capture the nuances of customer feedback and simulate a wider range of scenarios.
Moreover, future studies could examine the long-term effects of incorporating AI-generated feedback data into student learning. By conducting longitudinal studies, researchers could assess the sustainability of the skills gained through the relevant student learning and evaluate the impact on the learning results over time. Furthermore, given the rapid advancements in AI technology, future research could explore the integration of emerging AI capabilities, such as sentiment analysis. By harnessing these advanced AI tools, students could further enhance their ability to interpret and respond to customer feedback effectively, ultimately leading to more personalized and responsive service delivery.
Lastly, extending the scope of research beyond the café and restaurant to other sectors within the hospitality industry, such as hotels and catering services, could offer valuable comparative insights into the effectiveness of AI-driven training approaches across different contexts. By broadening the focus of inquiry, researchers can contribute to the development of best practices and guidelines for leveraging AI technologies to improve student learning performance in the broader hospitality sector.
6.7 Summary and Conclusion
In summary, this study presents AIRSim as an innovative tool to utilizing AI for students to practice their analysis skills in café and restaurant discipline. Our findings demonstrate the tool's capability in generating diverse simulated responses to user-uploaded questionnaires. While the study has its limitations, it offers valuable insights and opens avenues for future research in the domain of AI-driven student learning. Ultimately, AIRSim holds promise as a valuable tool for improving student learning in the hospitality industry.
Funding: Not applicable.
Conflicts of interest: There are no ethical issues or conflicts of interest emanating from this study.
Consent to Participate: This research does not involve human subjects. Therefore, consent to participate is not applicable.
Financial interests: The authors have no relevant financial or non-financial interests to disclose.
Data Availability: The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.
Code availability: Not applicable.
Authors' contributions: All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by the authors. All authors read and approved the final manuscript.
- Gray BJ, Matear SM, Matheson PK. Improving the performance of hospitality firms. Int J Contemp Hosp Manag. 2000;12:149–55. https://doi.org/10.1108/09596110010320643.
- Gamarra-Miranda MA, Rojas-Chávez J, Jauregui-Arroyo R, Rondon-Jara E. Relationship between Service Quality and Customer Satisfaction in Restaurants in Los Olivos, Peru. vol. 12, IEOM Society; 2022.
- Alhelalat JA, Habiballah MA, Twaissi NM. The impact of personal and functional aspects of restaurant employee service behaviour on customer satisfaction. Int J Hosp Manag. 2017;66:46–53. https://doi.org/10.1016/j.ijhm.2017.07.001.
- Oliveira ZC da C, Saldanha ES, Vong M. The Mediation Effect of Restaurant Image on the Relationship between Service & Food Quality and Customer Satisfaction. Timor Leste J Bus Manag. 2020;2:10–20. https://doi.org/10.51703/bm.v2i0.30.
- Kocaman EM, Türkmen BM. The Effects of Use of Restaurant Management Systems Perceived by the Personnel According to Their Demographic Characteristics. https://doi.org/10.4018/978-1-7998-9008-9.ch012
- Ishan MHARK, Chakma JD, Rozario RF. The Operational Impact of the Restaurants on Its Management System and Business Characteristics. Am J Multidiscip Res Innov. 2022;1:35–43.
- Ramnarayan, Joshi K, Reshi J, Memoria M, Gupta A, Rastogi N. AI/ML Based new smart customer feedback system for hospitality Industry. 2022 7th Int. Conf. Comput. Commun. Secur. ICCCS, 2022, p. 1–4. https://doi.org/10.1109/ICCCS55188.2022.10079480.
- Lee C, Hallak R, Sardeshmukh SR. Creativity and innovation in the restaurant sector: Supply-side processes and barriers to implementation. Tour Manag Perspect. 2019;31:54–62. https://doi.org/10.1016/j.tmp.2019.03.011.
- Brown TB, Mann B, Ryder N, Subbiah M, Kaplan J, Dhariwal P, et al. Language Models are Few-Shot Learners. 2020. https://doi.org/10.48550/arXiv.2005.14165.
- Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN, et al. Attention Is All You Need 2023. https://doi.org/10.48550/arXiv.1706.03762.
- Kaplan A, Haenlein M. Siri, Siri, in my hand: Who’s the fairest in the land? On the interpretations, illustrations, and implications of artificial intelligence. Bus Horiz. 2019;62:15–25. https://doi.org/10.1016/j.bushor.2018.08.004.
- Gao Z, Yi X, Sun M, Li W, Yang C, Liang J, et al. Jiuge: A Human-Machine Collaborative Chinese Classical Poetry Generation System. Proc. 57th Annu. Meet. Assoc. Comput. Linguist. Syst. Demonstr., Florence, Italy: Association for Computational Linguistics; 2019, p. 25–30. https://doi.org/10.18653/v1/P19-3005.
- Liu Y, Ott M, Goyal N, Du J, Joshi M, Chen D, et al. RoBERTa: A Robustly Optimized BERT Pretraining Approach 2019. https://doi.org/10.48550/arXiv.1907.11692.
- Lan Z, Chen M, Goodman S, Gimpel K, Sharma P, Soricut R. ALBERT: A Lite BERT for Self-supervised Learning of Language Representations 2020. https://doi.org/10.48550/arXiv.1909.11942.
- Wiese G, Weissenborn D, Neves M. Neural Domain Adaptation for Biomedical Question Answering 2017. https://doi.org/10.48550/arXiv.1706.03610.
- Chalkidis I, Fergadiotis M, Malakasiotis P, Aletras N, Androutsopoulos I. LEGAL-BERT: The Muppets straight out of Law School 2020. https://doi.org/10.48550/arXiv.2010.02559.
- Hendrycks D, Burns C, Basart S, Zou A, Mazeika M, Song D, et al. Measuring Massive Multitask Language Understanding 2021. https://doi.org/10.48550/arXiv.2009.03300.
- Li J, Luo Z, Huang H, Ding Z. Towards Knowledge-Based Tourism Chinese Question Answering System. Mathematics 2022;10:664. https://doi.org/10.3390/math10040664.
- Do P, Phan THV, Gupta BB. Developing a Vietnamese Tourism Question Answering System Using Knowledge Graph and Deep Learning. ACM Trans Asian Low-Resour Lang Inf Process 2021;20:81:1–81:18. https://doi.org/10.1145/3453651.
- Wang Y. Generative AI in Operational Risk Management: Harnessing the Future of Finance 2023. https://doi.org/10.2139/ssrn.4452504.
- Romanus N, Danielsson S. Question-Answering in the Financial Domain n.d.
- Biever C. ChatGPT broke the Turing test — the race is on for new ways to assess AI. Nature 2023;619:686–9. https://doi.org/10.1038/d41586-023-02361-7.
- Campello de Souza B, Serrano de Andrade Neto A, Roazzi A. Are the New AIs Smart Enough to Steal Your Job? IQ Scores for ChatGPT, Microsoft Bing, Google Bard and Quora Poe 2023. https://doi.org/10.2139/ssrn.4412505.
- Ji Y, Gong Y, Peng Y, Ni C, Sun P, Pan D, et al. Exploring ChatGPT’s Ability to Rank Content: A Preliminary Study on Consistency with Human Preferences 2023. https://doi.org/10.48550/arXiv.2303.07610.
- Orrù G, Piarulli A, Conversano C, Gemignani A. Human-like problem-solving abilities in large language models using ChatGPT. Front Artif Intell 2023;6. https://doi.org/10.3389/frai.2023.1199350.
- Leong K, Sung A. An exploratory study of how emotion tone presented in a message influences Artificial Intelligence (AI) powered recommendation system. J Technol Innov 2023;3:80–4. https://doi.org/10.26480/jtin.02.2023.80.84.
- Zhao Y, Huang Z, Seligman M, Peng K. Risk and prosocial behavioural cues elicit human-like response patterns from AI chatbots. Sci Rep 2024;14:7095. https://doi.org/10.1038/s41598-024-55949-y.
- Punar Özçelik N, Yangın Ekşi G. Cultivating writing skills: the role of ChatGPT as a learning assistant—a case study. Smart Learn Environ 2024;11:10. https://doi.org/10.1186/s40561-024-00296-8.
- Ibrahim A, Senthilkumar K, Saito K. Evaluating responses by ChatGPT to farmers’ questions on irrigated lowland rice cultivation in Nigeria. Sci Rep 2024;14:3407. https://doi.org/10.1038/s41598-024-53916-1.
- Haman M, Školník M, Lošťák M. AI dietician: Unveiling the accuracy of ChatGPT’s nutritional estimations. Nutrition 2024;119:112325. https://doi.org/10.1016/j.nut.2023.112325.
- Ali R, Connolly ID, Tang OY, Mirza FN, Johnston B, Abdulrazeq HF, et al. Bridging the literacy gap for surgical consents: an AI-human expert collaborative approach. Npj Digit Med 2024;7:1–6. https://doi.org/10.1038/s41746-024-01039-2.
- Koc E, Hatipoglu S, Kivrak O, Celik C, Koc K. Houston, we have a problem!: The use of ChatGPT in responding to customer complaints. Technol Soc 2023;74:102333. https://doi.org/10.1016/j.techsoc.2023.102333.
- Stampfl R, Ivkić I, Geyer B. Role-Playing Simulation Games using ChatGPT 2024. https://doi.org/10.48550/arXiv.2402.09161.
- Chen H, Chen H, Yan M, Xu W, Gao X, Shen W, et al. RoleInteract: Evaluating the Social Interaction of Role-Playing Agents 2024. https://doi.org/10.48550/arXiv.2403.13679.
- Maurya RK. A qualitative content analysis of ChatGPT’s client simulation role-play for practising counselling skills. Couns Psychother Res 2023. https://doi.org/10.1002/capr.12699.
Table 1 is available in the Supplementary Files section.
No competing interests reported.