Next Article in Journal
Identification of Emerging Technological Hotspots from a Multi-Source Information Perspective: Case Study on Blockchain Financial Technology
Next Article in Special Issue
A Training-Free Latent Diffusion Style Transfer Method
Previous Article in Journal
User Experience and Usability of Voice User Interfaces: A Systematic Literature Review
Previous Article in Special Issue
What Is Your Favorite Gender, MLM? Gender Bias Evaluation in Multilingual Masked Language Models
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Enhancing Task-Oriented Dialogue Systems through Synchronous Multi-Party Interaction and Multi-Group Virtual Simulation

by
Ellie S. Paek
*,
Talyn Fan
,
James D. Finch
and
Jinho D. Choi
*
Department of Computer Science, Emory University, Atlanta, GA 30322, USA
*
Authors to whom correspondence should be addressed.
Information 2024, 15(9), 580; https://doi.org/10.3390/info15090580
Submission received: 6 August 2024 / Revised: 11 September 2024 / Accepted: 12 September 2024 / Published: 19 September 2024
(This article belongs to the Special Issue Feature Papers in Artificial Intelligence 2024)

Abstract

:
This paper presents two innovative approaches: a synchronous multi-party dialogue system that engages in simultaneous interactions with multiple users, and multi-group simulations involving virtual user groups to evaluate the resilience of this system. Unlike most other chatbots that communicate with each user independently, our system facilitates information gathering from multiple users and executes 17 administrative tasks for group requests adeptly by leveraging a state machine-based framework for complete control over dialogue flow and a large language model (LLM) for robust context understanding. Assessing such a unique dialogue system poses challenges, as it requires many groups of users to interact with the system concurrently for an extended duration. To address this, we simulate various virtual groups using an LLM, each comprising 10–30 users who may belong to multiple groups, in order to evaluate the efficacy of our system; each user is assigned a persona and allowed to interact freely without scripts. As a result, our system shows average success rates of 87% for task completion and 89% for natural language understanding. Comparatively, our virtual simulation, which has an average success rate of 80%, is juxtaposed with a group of 15 human users, depicting similar task diversity and error trends. To our knowledge, it is the first work to show the LLM’s potential in both task execution and the simulation of a synchronous dialogue system to fully automate administrative tasks.

1. Introduction

Administrative tasks play a pivotal role in upholding organizational efficacy. Though they are typically overseen by human assistants, the increasing demand for swift task execution poses ongoing strains. To tackle this issue, chatbots have emerged as invaluable assets, streamlining tasks such as appointment scheduling and message dissemination [1,2,3]. Many dialogue systems aiming to tackle knowledge-based inquiries have also emerged, proving to be valuable assets within organizations [4,5,6,7,8,9]. Nonetheless, most current chatbots assist each user independently, despite many organizational processes requiring the assistant to engage with multiple users simultaneously to complete tasks. As a result, these chatbots often find themselves merely managing personal to-do lists and answering knowledge-based questions, leaving the bulk of the workload still for humans to handle.
Large language models (LLMs) can enable us to develop chatbots that facilitate synchronous multi-party dialogues, where multiple users are simultaneously engaged with a chatbot to fulfill their tasks. With LLMs’ notable enhancements, there is a growing interest in training LLMs to build end-to-end dialogue systems as well as in augmenting them to improve upon dialogue systems [7,8,10,11]. However, inherent challenges from LLMs, such as hallucination and lack of controllability [12,13], make deploying pure LLM-based systems difficult in practice. To incorporate LLMs’ language capabilities into real systems, there is a need to couple them with more manageable frameworks to ensure in-control task completion while achieving robust language interactions.
Developing a synchronous multi-party dialogue system is challenging, but evaluating such a system presents yet another hurdle. While a human-based evaluation offers the closest representation of real-life environments, recruiting a large group of users to thoroughly assess these dialogue systems over an extended period is time-consuming and resource-intensive. To address this, automated methods for simulating virtual users have been proposed; however, most of them simulate single users at a time, not simultaneous engagements from multiple users [14,15,16,17]. Thus, devising innovative solutions for conducting virtual simulations to assess such workload-intensive systems is imperative to develop dialogue systems that are well-prepared for real-life environments at minimal cost.
We present a dialogue system adept at conducting synchronous dialogues among multiple users for 17 administrative tasks (Section 3). In tandem, we introduce a multi-group simulator to facilitate the evaluation of our system by simulating users among multiple groups (Section 4). Our system is tested across various virtual environments produced by the simulator, showcasing high confidence in handling those tasks, and the results are promising for the effectiveness of our simulator and its resemblance to human interaction (Section 5). To our knowledge, this is the first system that comprehensively manages administrative tasks via synchronous multi-party dialogues and the first simulator accommodating users shared across multiple virtual groups.

2. Related Work

2.1. LLMs in Task-Oriented Dialogue Systems

With the emergence of LLMs for various natural language processing (NLP) tasks [18,19,20], their applications in task-oriented dialogue systems have grown, particularly for intent classification and dialogue state tracking. For intent classification, Zhang et al. [21] proposed a few-shot model using contrastive pre-training and fine-tuning, outperforming previous transformer-based models. Parikh et al. [22] adapted GPT for zero-shot prompting, showing competitive performance to few-shot prompting. For dialogue state tracking, Heck et al. [23] assessed GPT accomplishment of this task on two popular datasets [24,25], which showed state-of-the-art results. Feng et al. [26] developed a framework that leveraged open-source LLMs, yielding results comparable to those achieved with GPT.
While LLMs have been evaluated for those tasks, the research that has explored their sequential execution in dialogue systems is limited. Bang et al. [27] and Chung et al. [28] adapted GPT to build end-to-end task-oriented dialogue systems, surpassing prior systems in task execution. While such end-to-end systems offer many benefits, achieving fine-level controllability using these systems—a crucial aspect for real applications—proves challenging.
LLMs have also been individually employed and augmented to create systems capable of managing administrative tasks. For instance, frameworks presented by Gebreab et al. [4], Hays and White [5], and Musumeci et al. [6] illustrate how LLM augmentations are applied to administrative tasks across various domains, including cybersecurity and healthcare. These frameworks typically focus on information retrieval and extraction from user input [29]. Works such as Jung et al. [7] and Pan et al. [8] also developed task-oriented dialogue systems in which LLMs were used for knowledge-grounded augmentation. These systems outperformed most systems built on pre-trained models in knowledge-grounded administrative tasks, such as answering frequently asked questions [7]. However, most existing dialogue systems are not equipped to manage interpersonal administrative tasks, which require more intricate frameworks in conjunction with LLMs. Thus, we adopt a state machine-based framework to regulate dialogue flows, while utilizing an LLM for language understanding and generation.

2.2. Synchronous Multi-Party Chatbots

The domain of synchronous multi-party dialogue systems, especially those for task-oriented dialogues, is underexplored. Toxtli et al. [30] introduced the TaskBot, allowing users to request that other users be reminded to complete tasks. The TaskBot effectively facilitated the transfer of reminders between users, fostering asynchronous communication among them. Mendoza et al. [9] presented another chatbot for interactions between students and teachers (STBot), showcasing various features, including scheduling events for teachers, exchanging files among users, and addressing frequently asked questions.
While both chatbots exhibit multi-party dialogue capabilities, they rely on commercial frameworks; TaskBot operates in the Microsoft Bot Framework (Bot Framework: https://dev.botframework.com), and STBot is built on Google Dialogflow (Dialogflow: https://cloud.google.com/dialogflow). Hence, their functions are limited by the features offered in those frameworks, hindering the adaptation of LLMs. Our system stands out, as it enables real-time synchronous dialogues between the chatbot and users for general tasks without targeting specific groups, and it does so by fully leveraging LLMs.

2.3. Multi-Group Virtual Simulations

Due to the inefficiency in time and labor associated with assessing dialogue systems, several automated evaluation methods have been suggested, such as self-talk, where a chatbot engages in internal conversations by taking both the roles of the bot and users, or interactions between multiple different chatbots [31,32,33,34]. The advent of LLMs and their widespread application in diverse dialogue systems has further diversified evaluation approaches, integrating human, automated, and LLM-based assessments to evaluate generated responses [35,36]. Recently, Svikhnushina and Pu [37] introduced a dialogue evaluation pipeline employing an LLM to interact with chatbots on behalf of human users, and it showed promising results, although the LLM interaction was limited to one chatbot at a time.
LLMs have proven to be capable in replicating human-like behavior, and there have been prior works exploring the simulation of virtual users with various personalities using LLMs [38,39]. As a result of this capability, the use of LLMs for user simulation has been studied in task-oriented dialogue systems as well. Lu and Wang [17] developed a framework that generated virtual students using LLMs to evaluate the quality of multiple choice questions. Sekulic et al. [14], Wan et al. [15] employed LLMs fine-tuned on augmented datasets and compared them with existing models such as Llama [18] and GPT. Liu et al. [16] suggested applying multiple simulators, with each simulator behaving as a distinct user, instead of employing a single simulator to represent multiple users when training task-oriented dialogue systems. Despite the advancements in using LLMs to simulate virtual personalities, however, synchronous multi-party simulation is still underexplored, particularly in a multiple group setting where the same user can belong to several groups, as introduced in our work.

3. Assistant Bot

Our dialogue system, the Assistant Bot, operates on the Emora State Transition Dialogue Manager (STDM) framework, which offers versatile ways of defining dialogue states and utilizing third-party APIs for the development of an end-to-end dialogue system. Upon user input, STDM determines a relevant state by interpreting the input context, moves to another state based on this interpretation, and generates an output using a template [40]. While STDM supports a rapid method of building a dialogue system with great controllability, it has several limitations. First, STDM relies on its own formal grammar, NATEX (NATural language EXpression), for language understanding, which often falls short in accommodating the diverse expressions encountered during conversations. Second, each user’s database is stored independently, thus, information extracted from one user cannot be shared across other users; although this adds an extra layer of security, it complicates the process of conducting synchronous multi-party dialogues. Third, it was originally designed for passive interactions, meaning that a chatbot built on STDM waits for users to initiate conversations without proactively engaging with them.
Section 3.1, Section 3.2 and Section 3.3 address these three limitations in STDM to develop a more robust dialogue system. Moreover, Section 3.4 outlines the administrative tasks that the Assistant Bot can handle. Figure 1 showcases the framework and execution process of the Assistant Bot, and Figure 2 demonstrates an example of synchronous multi-party dialogue between two mock users, showcasing the functionalities of the Assistant Bot.

3.1. Language Understanding and Generation

To enhance natural language understanding (NLU), our system uses GPT-3.5 Turbo (henceforth, GPT), known for its exceptional performance, instead of NATEX [41,42,43,44].
For each user input, a few-shot GPT prompt template is employed, pairing a question prompt with the user’s input to extract the necessary information. GPT is directed to respond in a JSON schema, which is then integrated into the system’s framework as a set of variables that the Assistant Bot accesses for subsequent tasks. The few-shot GPT prompt template is as follows:
{full_request} Respond only in the JSON schema such as {examples}. Analyze this quote: “{quote_to_analyze}
In this template, full_request corresponds to the specific request GPT must fulfill in order to extract the relevant information, examples refers to example JSON schemas that the LLM should use as a response format, and quote_to_analyze refers to the user’s input. For instance, if the user provides the name of the recipient with whom they wish to schedule a meeting, the variables are populated with information pertinent to the task. The final GPT prompt is as follows:
The speaker wants to schedule a meeting with another person. What is that other personś name or ID? Ignore honorifics such as Dr, Mr, Mrs, Ms, or Mx. Respond only in the JSON schema such as {V.other_person_full_name.name: ‘Nora Allen’} or {V.other_person_full_name.name: ‘N/A’} if unavailable. Analyze this quote: “I need to meet with Dr. Smith”
Figure 3 illustrates how GPT is employed for NLU. For each user input, GPT infers the task (Make Appointment) and extracts relevant information (Professor Smith). If necessary, the extracted information is matched to database entries (Elizabeth Smith). When there is an ambiguity (e.g., multiple users with the last name Smith who are also professors), it enters an intermediate state for further clarification (e.g., Which of these people did you mean?). This information is then used to determine the next state, and an output is created by a template predefined for the state. Finally, the templated output is fed into GPT to diversify the final output. Should GPT fail to extract the necessary information, a combination of NATEX and regular expressions is employed as a backup to extract the information. Alternatively, the conversation flow is redirected to a state where the chatbot asks for clarification and reattempts to extract the required information.

3.2. Shared Database

For seamless information exchange and facilitation of synchronous communication and task execution, our system employs MongoDB, a NoSQL database system (MongoDB: https://www.mongodb.com). Each user record in the database contains a unique user ID, groups that the user belongs to, conversation history and relevant information interpreted by GPT (Section 3.1), and an inbox storing the user’s appointments, messages, feedback questions, and responses to groups. Access to the records of other users is strictly restricted for any user. When a user requests a task from another user (e.g., making an appointment), the user indicates their preferences (e.g., date and location), stored in the user’s record, and grants the system permission to share particular parts of the information with the other user. Only the permitted information is shared among the targeted users to accomplish the task, allowing our system to send requested messages, anonymously or with names attached, to individuals or specific groups.

3.3. Proactive Engagement

Many administrative tasks require the system to actively reach out to users in a timely manner, which is difficult to accomplish without a proper interface. For proactive engagement with our users, a mobile app is created using Flutter, an open-source framework for developing multi-platform applications (Flutter: https://flutter.dev). Figure 4 portrays the dashboard and chat interface of the Assistant Bot application. When a user requests a task, the app sends a push notification to the other user’s application on their mobile device, facilitating engagement with our system among human users (Section 5) (our app is available on both iOS App Store and Google Play).

3.4. Administrative Tasks

Our system manages four types of administrative tasks, as outlined in Table 1. Each type of task corresponds to an individual component within the Assistant Bot which the chatbot transitions to if the identified request corresponds to the task listed in Table 1.

3.4.1. Names and Hub

Upon the initial interaction, the Assistant Bot requests the user’s official name and preferred names. While the official name is stored in the database and matched for task requests from other users, our system dynamically incorporates preferred names into subsequent responses to enhance user engagement. Upon name confirmation, it moves to the hub state, where it allows the user to request any task, categorizes it into a task in Table 1, and directs the user to the corresponding state.

3.4.2. Appointments

The Assistant Bot streamlines appointment scheduling by managing details such as date, time, location, and purpose. Upon mutual confirmation, it updates users’ schedules and resolves any scheduling conflicts. Requesting users can edit or cancel pending appointment requests, while both requesting and requested users can reschedule any appointments with passed or conflicting dates.

3.4.3. Messages

A user can dispatch messages to any other users in the database via the Assistant Bot. Users can send messages either anonymously or with their names; however, anonymity is maintained externally to ensure user safety.

3.4.4. Groups

The Assistant Bot allows users to create groups, with the creator becoming the leader and being able to add other users as members. Groups are identified by unique IDs, and users can check the groups they belong to at any time. Leaders oversee group activities, including creation, deletion, and communication through announcements. They can also request feedback from group members and review statistics on member responses. Members can respond to feedback requests from the leader and leave groups at their discretion; however, rejoining requires the leader’s permission.

4. Multi-Group Virtual Simulator

It is challenging to coordinate with human users to assess the Assistant Bot’s ability to handle multiple users in different groups, especially for researchers lacking established user bases. Thus, we present a multi-user simulator harnessing GPT’s capabilities in persona impersonation and dialogue generation [45,46]. Each user is personalized with a distinct prompt and their past dialogue history with the chatbot. The rest of this section outlines our simulation process.

4.1. Preprocessing

Prior to simulation, personas for u users across g groups are first generated by GPT. Each persona contains the user’s unique ID, name, occupation, personality, and associated class IDs. Leveraging GPT for persona generation enhances scalability, as increasing u and g requires no extra human effort with a small additional cost.

4.2. Initialization

Each virtual simulation starts with initialization, when a user greets the Assistant Bot. The Assistant Bot then asks the user for their name. Once our system receives the name, it transitions the user to the hub state, so the user can commence a simulated conversation. This process is applied to designated users until they are all initialized.

4.3. Interactions

During interactions, the simulator dynamically loads the user’s dialogue history into a prompt template, which is then fed into GPT to generate an output tailored to the user’s persona.
The simulator presents this output as a user turn and generates a subsequent system turn according the corresponding state. The conversation history, including the last system output to that user, gets updated accordingly.
Upon each system turn, two additional functions are executed using GPT. The first monitors the ongoing conversation between the current user and the chatbot, checking for any pending notifications or tasks that have been resolved. The results, presented as a dictionary containing notification and task counts for each user, are used to update the relevant users’ notifications and tasks in the database. The second function ascertains whether or not the conversation is concluded. If not, it continues with the current user until it meets a natural conclusion.
A conversation concludes once all relevant states have been explored to address the requested task. When a conversation with one user concludes, the simulator selects a different user based on their task and notification status, as follows:
argmax i U ( ( α · n i ) + ( β · t i ) Σ j U ( α · n j ) + ( β · t j ) )
n i and t i are the number of notifications and tasks of the i’th user, respectively. U is the set of all user numbers, excluding the one assigned to the current user. In our case, α = 2.5 and β = 1 are used such that users with higher notification counts receive greater weight, simulating real-life behavior, where people tend to prioritize notifications. From users with the highest weights, the next user is selected via random sampling to continue the simulation.
The prompts for generating the simulated user’s response, as well as those for updating notifications and tasks and checking if a conversation has concluded, are illustrated in Appendix A.1.1.

4.4. Termination

Each simulation is constrained by a maximum number of turns m, encompassing both system and user turns. The simulation terminates either upon reaching the turn limit or resolving all tasks and notifications for the designated users.

5. Experiments

5.1. Virtual Simulation

To assess the efficacy of our system, various college class scenarios are simulated. Given ( u , g ) , where u / g equates to the respective number of users/groups (Section 4), five rounds of simulation are conducted, each composing different combinations, as follows:
( u , g ) { ( 10 , 1 ) , ( 20 , 1 ) , ( 20 , 2 ) , ( 20 , 3 ) , ( 30 , 1 ) }
For instance, in the scenario ( 20 , 2 ) , we simulate 20 students across 2 classes, ensuring every student takes at least one of those two classes. Additionally, one instructor is simulated for each class. Table A1 in Appendix A.1 describes the list of personas generated by GPT during the preprocessing stage (Section 4). Moreover, the maximum number of turns for termination in each round is set to m = 500 (Section 4).

5.2. Human Simulation

To compare the behaviors of the virtual classroom with those of humans, a group comprising one professor and 15 college and graduate students is recruited to simulate a classroom scenario, interacting with our Assistant Bot via the mobile app (Section 3.3). Each user is provided a persona description to role-play, in tandem with those assigned to the virtual users in the simulation (Table A1 in Appendix A.1). To ensure fair comparisons, the class details for both the virtual and human simulations are kept consistent. This simulation comprises a total of 1821 turns.
We acknowledge that the sample size in the experiments is relatively small, which may limit the generalizability of the results. Future work will involve conducting experiments with larger sample sizes to ensure broader generalizability and to further validate the effectiveness of the virtual simulation and Assistant Bot.

5.3. Evaluation Metrics

For each round of simulation, the Assistant Bot and the virtual simulator are evaluated and compared to the human simulation. The task completion of the Assistant Bot is tested by how successfully it handles each requested task using the metrics below:
  • NLU: The failure rate due to NLU shortcomings (in %).
  • STD: The failure rate due to inherent limitations in the STDM framework, hindering it from exiting the current dialogue state until it is fully resolved, even when the user requests another task (in %).
  • SUC: The overall success rate; 100 − (NLU + STD).
  • SWS: The success rate without accounting for STD, used to evaluate system performance excluding fails due to limitations of the STDM framework; 100 × SUC/(100 − STD).
Moreover, each turn generated by the Assistant Bot is assessed for its robustness in natural language understanding using the following metrics:
  • INT: The error rate resulting from inaccurate intent classification by GPT (in %).
  • DST: The error rate resulting from inaccurate dialogue state tracking by GPT (in %).
  • STD: The error rate due to the STDM limitations, leading to NLU failures (in %).
  • SUC: The overall success rate; 100 − (INT + DST + STD).
  • SWS: The success rate without considering STD; 100 × SUC/(100 −STD).
Finally, the appropriateness rate (APP) of the simulator is assessed for each virtual user turn, which addresses various factors such as hallucinations by GPT, responses irrelevant to the Assistant Bot’s last response, and repetitive statements disregarding the user’s previous responses. If the repetition pertains to the requested task, it is still considered relevant.

5.4. Results

Table 2 presents the evaluation results for our Assistant Bot and Virtual Simulator. For task completion, performance is notably influenced by STDM; when discounting STDM errors, it achieves high success rates of 87.1% and 81.6% for the virtual and human simulations, respectively. Interestingly, comparisons of different class numbers, ( u = 20 , g { 1 , 2 , 3 } ) , show improved SWS with more classes; distributing a fixed number of students across a greater number of classes reduces the size of each class, which may lead to more successful task completion.
For natural language understanding, the average success rates rise by 19.6% and 22.7% for the virtual and human simulations, respectively, when discarding the STDM errors, suggesting that while STDM enables rapid prototyping of dialogue systems, a more robust framework is needed for real applications. Both SUC and SWS remain relatively stable across all scenarios, confirming the robustness of the NLU approach enabled by GPT in our system.
Comparing the virtual and human simulations reveals similar performance trends, underscoring the adaptability of our virtual simulation for evaluating synchronous dialogue system performance. Both simulations produce comparable numbers of tasks per turn; 44% and 40% of turns in the virtual and human simulations, respectively, are task requests. While the human simulation comprises a broader range of tasks (20) compared to the virtual one (14), the virtual simulation still elicits a sufficiently diverse task distribution, facilitating effective evaluation of the bot’s capabilities. It is important to note, however, that it is easy to design a virtual simulation that accommodates a dynamic range of tasks, which can result in a more varied task distribution, as users are completely controllable in this setting.

6. Discussion

6.1. Error Analysis on the Assistant Bot

For each erroneous turn of the Assistant Bot during the virtual simulation, the generated dialogues and preceding context are examined to understand its limitations. Table 3 highlights the common errors that resulted in incorrect responses from the Assistant Bot. Examples of these errors are in Appendix A.3.

6.1.1. STDM Limitations

The primary source of the Assistant Bot’s errors stems from limitations in STDM, leading to various issues in dialogue flow. Of these limitations, g errors emerge as prominent issues, as the Assistant Bot cannot move on to a new task as a result of the STDM locking it into the current state until fully resolved, even when the user wishes to abandon the current task and transition to the new one (Figure A10). Although our system mainly uses GPT for NLU, it still defaults to NATEX as a fallback NLU method in cases where GPT fails due to internet connection or server issues. NATEX errors arise when NATEX takes over GPT for NLU with incorrect interpretation (Figure A11). Certain errors are induced by unexpected system bugs in STDM (Figure A12), while a minor portion of errors are attributed to STDM’s inability to handle multiple task requests in a single turn, in which cases, STDM processes one of the requests and disregards the others (Figure A13).

6.1.2. NLU Errors in Dialogue State Tracking (DST)

Most of the NLU errors occur due to GPT shortcomings in dialogue state tracking (DST) and intent classification (IC) tasks. The primary cause of DST errors is GPT’s limited ability to robustly extract multiple dialogue states at once, as the GPT prompt becomes too complicated when many states are included (Figure A14). Additionally, GPT variability and inconsistency sometimes result in incorrect information extraction.

6.1.3. NLU Errors in Intent Classification (IC)

For IC, the main source of errors stems from context management issues, where user inputs are misinterpreted as certain tasks. Due to overlap among tasks within the same type, such as making and canceling appointments, context ambiguity can result in incorrect intent classification (Figure A15). In addition, GPT may wrongly classify the user’s intended task, causing the Assistant Bot to enter an incorrect state.

6.2. Error Analysis on the Virtual Simulation

Every erroneous turn from the virtual simulation is analyzed to assess its effectiveness in handling synchronous multi-party dialogues. Table 4 shows the causes of these inappropriate responses. The most frequent errors occur when the virtual simulation ignores contexts from the Assistant Bot’s previous response (Figure A16). While these errors disrupt dialogue flow, it is important to note that they are not unlike human behavior, as humans often respond in ways that overlook the chatbot’s previous response.
Another primary cause stems from the LLM’s inability to focus on the provided information to accurately simulate the user. The simulated user frequently forgets parts of the conversation history, leading to requests for tasks that have already been completed (Figure A17). Furthermore, it fails to recall the provided information, generating irrelevant or incorrect content in relation to the given context.
The third type of error occurs when the simulation forgets information about the user, resulting in content that is inconsistent with the simulated user’s information (Figure A18). A small portion of errors occurs when the simulation becomes confused about whom it is conversing with, leading to responses that address the Assistant Bot incorrectly. This often happens when the bot delivers a message from another (simulated) user, causing the simulation to address the user rather than the Assistant Bot (Figure A19). Finally, role reversal occurs when the simulation fails to act as the user, instead generating responses that resemble the characteristics of the Assistant Bot (Figure A20).

6.3. Limitations

The limitations encountered by the Assistant Bot are largely rooted in the inherent constraints of the STDM framework. One significant issue is the use of NATEX. While NATEX is effective for the rapid prototyping of conversation transitions, it falls short in terms of natural language understanding due to its manually constructed nature. Although NATEX can be expanded to handle a variety of user inputs, creating a comprehensive dictionary that captures all possible transitions is impractical. This challenge, combined with the inherent ambiguity of natural language, often leads to categorization and transition errors. Furthermore, state lock errors prevent the Assistant Bot from smoothly handling multiple tasks. This is particularly problematic when users request multiple tasks within the same utterance or when they wish to switch tasks mid-conversation.
Given these limitations, exploring an alternative framework to replace the STDM could provide significant improvements. Since the STDM is robust in generating relevant natural language, a framework that can replicate the STDM’s robust natural language generation while also dynamically navigating between different task-oriented conversation states is crucial for enhancing our chatbot’s performance.
Additionally, the limitations of GPT contribute to errors in both the Assistant Bot and the virtual simulation. For the Assistant Bot, GPT’s variability and its limited understanding of administrative task categorization can lead to incorrect information extraction, resulting in errors in dialogue state tracking and intent classification. For the virtual simulation, maintaining in-character responses that aligned with user profiles and conversation history proved challenging, especially when dealing with extensive context. Fine-tuning user prompts or the model specifically for simulation tasks could improve accuracy in these areas. Furthermore, as GPT-3.5 is currently used for both the Assistant Bot and the virtual simulation, utilizing more powerful models such as GPT-4 could enhance the performance of both the simulation responses and the Assistant Bot’s natural language understanding.

7. Conclusions

While addressing the complexities of synchronous multi-party dialogue and multi-group virtual simulation presents challenges, this work is the first to demonstrate promising potential of a task-oriented dialogue system in simultaneously handling multiple users and leveraging large language models (LLMs) to supplement human evaluation. Despite limitations in the STDM framework and variabilities in GPT, our system exhibited robust natural language understanding and task execution, proving its practicality as a real-world application. Our virtual simulation also demonstrated a diversity in task requests similar to human interactions, underscoring its reliability as an evaluative tool to overcome the challenge of recruiting human groups.
However, the limitations of the STDM framework, coupled with the inherent variability of GPT models, suggests several directions for future research. Future work should focus on refining context management to further improve simulation performance and reliability. Enhancing the Assistant Bot’s ability to understand situational nuances could also strengthen its natural language comprehension. Exploring alternative frameworks that offer stronger language generation capabilities, while addressing the constraints of the current STDM approach, could further advance the system’s effectiveness. Additionally, applying the virtual simulation methodology to a wider range of group settings beyond the context of college classrooms could help generalize its applicability and adaptability.
While the Assistant Bot and virtual simulation help make significant advancements in the automation of administrative tasks and multi-user dialogue management using LLMs, continued research and refinement are necessary to fully realize the potential of these innovative approaches in diverse and dynamic environments. We look forward to advancements in task-oriented dialogue systems that employ synchronous multi-user features, as well as in evaluation frameworks that leverage the robust capabilities of large language models.

Author Contributions

Conceptualization, J.D.C.; methodology, J.D.C. and J.D.F.; software, E.S.P., T.F., and J.D.C.; validation, E.S.P., T.F., J.D.F., and J.D.C.; formal analysis, E.S.P.; investigation, E.S.P., J.D.C., and J.D.F.; resources, J.D.C.; data curation, E.S.P. and T.F.; writing—original draft preparation, E.S.P. and J.D.C.; writing—review and editing, E.S.P. and J.D.C.; visualization, E.S.P. and J.D.C.; supervision, J.D.C. and J.D.F.; project administration, J.D.C.; funding acquisition, J.D.C. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by Emory University.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The original contributions presented in the study are included in the article, further inquiries can be directed to the corresponding authors.

Acknowledgments

We gratefully acknowledge the support of Emory University. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of Emory University.

Conflicts of Interest

The authors declare no conflicts of interest.

Appendix A

Appendix A.1. Virtual Simulation Prompts and Personas

Table A1. Personas given to the virtual users in the simulation. The first group of rows lists the instructors of each course (course written in Class), and the second group of rows lists the students. CS: Introduction to Computer Science, PSY: Introduction to Psychology, HIS: United States History: From Colonial Period to Present.
Table A1. Personas given to the virtual users in the simulation. The first group of rows lists the instructors of each course (course written in Class), and the second group of rows lists the students. CS: Introduction to Computer Science, PSY: Introduction to Psychology, HIS: United States History: From Colonial Period to Present.
NameClassPersonalityTendencies
Joseph ChoCSenthusiasticorganized and plans material meticulously
Amanda JohnsonPSYapproachableinteractive and encourages participation
Michael BrownHISanalyticalenjoys solving complex problems, often takes a methodical approach
Olivia AdamsCScreativeapproaches problems with unconventional thinking
Leila AhmedCScreativeenjoys experimenting, thinks outside the box
Mohammed AliCS, PSYanalyticalenjoys solving challenging problems
Michael BrownCSanalyticalenjoys solving complex problems, often takes a methodical approach
Valerie ChenCS, HISenthusiasticorganized and plans study schedules meticulously
Diego FernandezCSoutgoingprefers group study sessions, actively participates in class discussions
Amelia GarciaCS, PSY, HISindependentprefers to work alone, excels in self-paced learning
Rohan GuptaCS, PSY, HISanalyticalapproaches problems logically
Mary JohnsonCSintrovertedenjoys individual projects, excels in structured assignments
Fatima KhanCS, PSY, HIScreativeapplies creative problem-solving techniques
Daniel LeeCSteam-orientedworks well in groups, enjoys collaborative projects
Maria LopezCS, PSY, HIScuriouslikes exploring new concepts, often conducts independent research
Rachel MartinezCSoutgoingworks best in groups, enjoys discussing ideas with classmates
Ethan MillerCSdiligentthrives in structured learning environments, maintains organized notes
Yuki NakamuraCSfocusedattentive in lectures
Sophia NguyenCScuriousexplores additional resources beyond lecture material
Aisha PatelCS, PSYreservedprefers self-study, takes detailed notes during lectures
Aaliyah PatelCS, HISfriendlyenjoys group discussions
Nadia RamirezCS, PSY, HISorganizedcreates detailed study guides, follows a structured study routine
Emily RodriguezCSfriendlyprefers group study sessions, good at explaining concepts to peers
Isabella RodriguezCS, HISenthusiasticeager to learn, actively participates in class discussions
Elijah ScottCSpersistentconsistently follows through on projects, enjoys mastering difficult concepts
Alexandre SilvaCSintrospectiveprefers self-paced learning
Kumar SinghCS, HISenthusiasticeager learner, actively participates in group discussions
Ben SmithCS, PSYeasygoingprefers studying in a group, collaborates on group projects
Haruto SuzukiCS, HISreservedprefers studying alone, methodical in approach
Sato TanakaCS, PSYfriendlyprefers collaborative learning
Ryne WatersCSanalyticalworks well under pressure, completes assignments last minute
Chloe WhiteCScuriousenjoys exploring new topics
James WilsonCSadventurousenjoys hands-on projects and experimentation

Appendix A.1.1. Student and Professor Response Generation Prompts

The prompt used for generating student simulation responses is as follows:
You are {name}, a student in the {sim_class} class. You are {personality}, with studying tendencies of {studying_tendencies}. Online, you tend to {online_tendencies}. Your professor is {professor}, and your classmates are {classmates}. The class history so far is as such: {class_history}. This week, the material covered was {current_material}.
You have {num_notifs} notifications you MUST check through the bot first. After the bot APPROPRIATELY RESPONDS, you want to complete this task: {todo}.
You are talking to an assistant chatbot. Its functionalities are ONLY as follows:
- makes appointments with other students or professor
- sends messages to people (anonymized or not anonymized) to other students or professor
- create and delete groups/classes with leaders and members
- leave a group if you are a MEMBER and in a GROUP
- check what groups you are in
- make announcements within groups if you are a LEADER and in a GROUP
You will be given a conversation history and the assistant chatbot’s last response. Using the information and the given identity and personality, respond IN CHARACTER and APPROPRIATELY to the assistant chatbot’s last response. Pay attention to what the bot is asking for and carefully consider your own tasks that you want to get done. Respond as if YOU are the one typing and asking for assistance; do NOT add role-playing details such as “John Doe: hi” or “user: hi” or “how can I assist you today?”. In addition, do NOT REPEAT yourself, such as repeating “Thank you” or “add all students”. If the conversation has ended, consider asking for any updates if you received notifications. If it is not, respond with something DIFFERENT. If the bot is repeating itself, CANCEL the current functionality the bot is in to return to the hub state.
AN EXAMPLE of how it should be responded is provided below:
user_prompt =
History:
user: hi
Using the conversation history provided above as context, respond in character to the chatbot’s last response below:
bot: Before we begin, please provide your full name.
CORRECT RESPONSE:
John Smith
AN EXAMPLE of how it should be responded is provided below:
user_prompt =
History:
user: hi
bot: Before we begin, please provide your full name.
user: Gwen Stacy
bot: Thank you Gwen Stacy. Welcome! How should I call you?
user: Gwen
Using the conversation history provided above as context, respond in character to the chatbot’s last response below:
bot: Nice to meet you, Gwen. Anything I can do for you, Gwen?
CORRECT RESPONSE:
any updates?
ANOTHER CORRECT RESPONSE:
actually call me Stacy
AN EXAMPLE of how it should be responded is provided below:
user_prompt =
user: Can you help me schedule a study group session with my classmates for this week?
Using the conversation history provided above as context, respond in character to the chatbot’s last response below:
bot: Sure, I can help you schedule an appointment with someone. What is their full name?
CORRECT RESPONSE:
let’s see if David is free
ANOTHER CORRECT RESPONSE:
Never mind
AN EXAMPLE of how it should be responded is provided below:
user_prompt =
user: Call me Jasmine. Any updates?
Using the conversation history provided above as context, respond in character to the chatbot’s last response below:
bot: Nice to meet you, Jasmine. So, how’s it going, Jasmine?
CORRECT RESPONSE:
so any mail
ANOTHER CORRECT RESPONSE:
I want to send a message to Dr Yin about a bug in the hw
AN EXAMPLE of how it should be responded is provided below:
user_prompt =
user: You can call me Fatima. Thank you.
bot: Nice to meet you, Fatima. So what’s up, Fatima?
user: I wanted to ask if there’s a way for me to find a classmate to work together on the Data Structures homework?
bot: You’re not in any groups right now, or you do not have any groups that you can request feedback from!!
user: Actually, I wanted to ask if there’s a way for me to find a classmate to work together on the Data Structures homework. Can you help me with that?
Using the conversation history provided above as context, respond in character to the chatbot’s last response below:
bot: Do you also still want help with check the members in a group?
CORRECT RESPONSE:
no, it’s ok
AN EXAMPLE of how it should be responded is provided below:
bot: Understood, I’ll move on. Anything I can do for you, Joseph?
user: Any updates?
bot: There are no new messages in your inbox. Anything I can do for you, Joseph?
user: No, I’m all set for now. Thank you.
Using the conversation history provided above as context, respond in character to the chatbot’s last response below:
bot: You have 1 message in your appointment inbox. Would you like to respond to it now?
CORRECT RESPONSE:
yes
History:
{history}
Using the conversation history provided above and the characterś personality and class details as context, respond IN CHARACTER TO the chatbotś last response below:
Chatbotś Last Response:
{bot_last_response}
The prompt used for generating professor simulation responses is as follows:
You are {name}, a professor teaching the {sim_class} class. You are {personality}, with studying tendencies of {teaching_tendencies}. Online, you tend to {online_tendencies}.
Your students are {students}. The class history so far is as such: {class_history}. This week, the material covered was {current_material}.
You have {num_notifs} notifications you MUST check through the bot first. After the bot APPROPRIATELY RESPONDS, you want to complete this task: {todo}.
You are talking to an assistant chatbot. Its functionalities are ONLY as follows:
- makes appointments with other students or professor
- sends messages to people (anonymized or not anonymized) to other students or professor
- create and delete groups/classes with leaders and members
- leave a group if you are a MEMBER and in a GROUP
- check what groups you are in
- make announcements within groups if you are a LEADER and in a GROUP
You will be given a conversation history and the assistant chatbot’s last response. Using the information and the given identity and personality, respond IN CHARACTER and APPROPRIATELY to the assistant chatbot’s last response. Pay attention to what the bot is asking for and carefully consider your own tasks that you want to get done. Respond as if YOU are the one typing and asking for assistance; do NOT add role-playing details such as “John Doe: hi” or “user: hi” or “how can I assist you today?”. In addition, do NOT REPEAT yourself, such as repeating “Thank you” or “add all students”. If the conversation has ended, consider asking for any updates if you received notifications. If it is not, respond with something DIFFERENT. If the bot is repeating itself, CANCEL the current functionality the bot is in to return to the hub state.
AN EXAMPLE of how it should be responded is provided below:
user_prompt =
History:
user: hi
Using the conversation history provided above as context, respond in character to the chatbot’s last response below:
bot: Before we begin, please provide your full name.
CORRECT RESPONSE:
John Smith
AN EXAMPLE of how it should be responded is provided below:
user_prompt =
History:
user: hi
bot: Before we begin, please provide your full name.
user: Gwen Stacy
bot: Thank you Gwen Stacy. Welcome! How should I call you?
user: Gwen
Using the conversation history provided above as context, respond in character to the chatbot’s last response below:
bot: Nice to meet you, Gwen. Anything I can do for you, Gwen?
CORRECT RESPONSE:
any updates?
ANOTHER CORRECT RESPONSE:
actually call me Stacy
AN EXAMPLE of how it should be responded is provided below:
user_prompt =
user: Can you help me schedule a study group session with my classmates for this week?
Using the conversation history provided above as context, respond in character to the chatbot’s last response below:
bot: Sure, I can help you schedule an appointment with someone. What is their full name?
CORRECT RESPONSE:
let’s see if David is free
ANOTHER CORRECT RESPONSE:
Never mind
AN EXAMPLE of how it should be responded is provided below:
user_prompt =
user: Call me Jasmine. Any updates?
Using the conversation history provided above as context, respond in character to the chatbot’s last response below:
bot: Nice to meet you, Jasmine. So, how’s it going, Jasmine?
CORRECT RESPONSE:
so any mail
ANOTHER CORRECT RESPONSE:
I want to send a message to Dr Yin about a bug in the hw
AN EXAMPLE of how it should be responded is provided below:
user_prompt =
user: You can call me Fatima. Thank you.
bot: Nice to meet you, Fatima. So what’s up, Fatima?
user: I wanted to ask if there’s a way for me to find a classmate to work together on the Data Structures homework?
bot: You’re not in any groups right now, or you do not have any groups that you can request feedback from!!
user: Actually, I wanted to ask if there’s a way for me to find a classmate to work together on the Data Structures homework. Can you help me with that?
Using the conversation history provided above as context, respond in character to the chatbot’s last response below:
bot: Do you also still want help with check the members in a group?
CORRECT RESPONSE:
no, it’s ok
AN EXAMPLE of how it should be responded is provided below:
bot: Understood, I’ll move on. Anything I can do for you, Joseph?
user: Any updates?
bot: There are no new messages in your inbox. Anything I can do for you, Joseph?
user: No, I’m all set for now. Thank you.
Using the conversation history provided above as context, respond in character to the chatbot’s last response below:
bot: You have 1 message in your appointment inbox. Would you like to respond to it now?
CORRECT RESPONSE:
yes
History:
{history}
Using the conversation history provided above and the characterś personality and class details as context, respond IN CHARACTER TO the chatbotś last response below:
Chatbotś Last Response:
{bot_last_response}
For both the student and professor prompts, the brackets {} are replaced with the information provided in the selected user profile. Each label corresponds to the following information:
  • name: Name of user
  • sim_class: Classes that the user is part of
  • personality: Personality of the user
  • studying_tendencies: Studying tendencies of the user
  • online_tendencies: Online tendencies of the user
  • professor: If the user is a student, professors that user has for classes
  • classmates: If the user is a student, classmates that user has in classes
  • students: If the user is a professor, students that user has in class
  • class_history: History of what material was taught and what the user did in the past weeks of classes
  • current_material: What material is being taught at the point in time in which the user is conversing with the Assistant Bot
  • num_notifs: Number of notifications the user has
  • todo: Task(s) that user seeks to complete through Assistant Bot
  • history: Chat history with Assistant Bot.
  • bot_last_response: Assistant Bot’s last response

Appendix A.1.2. Tasks and Notifications Prompt

The prompt used for determining the number of tasks and notifications for a given user is as follows:
You will be given a conversation history between a user and an assistant chatbot.
Using it and the information provided, determine if there are any notifications are NOT resolved and tasks that are NOT completed. A notification is ONLY resolved when the user has completely received the message that the bot is delivering. A task is ONLY completed when the bot has completed the task the user requests and concludes as such. Respond ONLY in a dictionary format such as {‘notifications’ : 1, ‘tasks’: 1}, or ‘N/A’ if unable to be determined.
An EXAMPLE of how it should be responded is provided below:
user_prompt =
This user has received 0 notification and their task is as follows: send a message to Gwen about studying together. The number of notifications and tasks is presented in a dictionary format below:
{‘notifications’: 0, ’tasks’ : 1}
History:
— smith is now talking —
user: hi
bot: Before we begin, please provide your full name.
user: Gabriel Smith
bot: Thank you Gabriel Smith. Welcome! How should I call you?
user: How about Darth Vader
bot: Nice to meet you, Darth Vader. How can I assist you, Darth Vader?
Based on the history above, how many remaining tasks and notifications does this user have?
CORRECT RESPONSE:
{‘notifications’: 0, ‘tasks’ : 1}
An EXAMPLE of how it should be responded is provided below:
user_prompt =
This user has received 2 notifications and their task is as follows: send a message to Ben about studying together. The number of notifications and tasks is presented in a dictionary format below:
{‘notifications’: 2, ‘tasks’ : 1}
History:
— fan is now talking —
user: You can call me Gwen.
bot: Nice to meet you, Gwen. What do you need next, Gwen?
user: I need to send a message to Ben about studying together.
bot: Okay! Would you like to send this message anonymously?
user: No, I don’t need to send the message anonymously.
bot: Alright! Can you please tell me the message that you want to send to Ben Smith?
user: Sure! Here’s the message I want to send to Ben: “Hey Ben, I was wondering if you’d be interested in studying together for our Intro to CS class. I find that collaborating with someone helps me understand the material better. Let me know if you’re available and we can set up a study session. Thanks!”
bot: Thank you! I’ll send off your message.
Based on the history above, how many remaining tasks and notifications does this user have?
CORRECT RESPONSE:
{‘notifications’: 2, ‘tasks’ : 0}
An EXAMPLE of how it should be responded is provided below:
user_prompt =
This user has received 2 notifications and their task is as follows: send a message to Dr. Cho about bug. The number of notifications and tasks is presented in a dictionary format below:
{‘notifications’: 2, ‘tasks’ : 1}
History:
— ali is now talking —
user: I need to send a message to Dr. Cho.
bot: Okay! Would you like to send this message anonymously?
user: No, I don’t need to send the message anonymously.
bot: Alright! Can you please tell me the message that you want to send to Ben Smith?
user: Subject: Possible Bug in Homework Assignment
Hi Dr. Cho,
I hope you’re doing well. I wanted to bring to your attention a possible bug that I’ve noticed in the current homework assignment. After reviewing the requirements, I believe there may be an issue with the expected output for a specific problem. I have detailed my findings and suggestions for resolving this in the document attached to this message.
Thank you for your attention to this matter.
Best regards,
Muhammad Ali
bot: Welcome! How should I call you?
Based on the history above, how many remaining tasks and notifications does this user have?
CORRECT RESPONSE:
{‘notifications’: 2, ‘tasks’ : 1}
An EXAMPLE of how it should be responded is provided below:
user_prompt =
This user has received 1 notifications and their task is as follows: create a group for school project with Daniel and Han. The number of notifications and tasks is presented in a dictionary format below:
{‘notifications’: 1, ‘tasks’ : 1}
History:
— tenali is now talking —
user: any updates?
bot: You have 1 message in your appointment inbox. Would you like to respond to it now?
user: yes
Based on the history above, how many remaining tasks and notifications does this user have?
CORRECT RESPONSE:
{‘notifications’: 1, ‘tasks’ : 1}
An EXAMPLE of how it should be responded is provided below:
user_prompt =
This user has received 0 notifications and their task is as follows: make an appointment with Professor Johnson. The number of notifications and tasks is presented in a dictionary format below:
{‘notifications’: 0, ‘tasks’ : 1}
History:
— patel is now talking —
user: You can call me Aisha.
bot: Nice to meet you, Aisha. What can I do for you, Aisha?
user: I would like to make an appointment with Professor Johnson for help with hash tables.
bot: Okay! Please tell me your first choice for a meeting day and time with Amanda Johnson.
user: I would prefer to meet with Professor Johnson on Thursday afternoon around 3:00 PM.
bot: Ok, I think I got it. Is this what you meant?
03/07/2024 (03:00 PM) to 03/07/2024 (03:30 PM)
user: Yes, that’s correct. Thank you for scheduling the appointment with Professor Johnson on Thursday afternoon from 3:00 PM to 3:30 PM.
bot: Awesome, I’ll see if they are free. Furthermore, where do you want to meet with Amanda Johnson?
Based on the history above, how many remaining tasks and notifications does this user have?
CORRECT RESPONSE:
{‘notifications’: 0, ‘tasks’ : 1}
This user has received {notifs} notifications and their task is as follows: {todo}. The number of notifications and tasks is presented in a dictionary format below:
{user_notif_data}
History:
{history}
Based on the history above, how many remaining tasks and notifications does this user have?
Each label corresponds to the following information from the selected user profile:
  • notifs: Number of notifications the user has
  • todo: Task(s) that user seeks to complete through Assistant Bot
  • user_notif_data: The user’s number of notifications and tasks to complete, in dictionary format. {‘notifications’: the number of notifications, ‘tasks’: the number of tasks}
  • history: Chat history with Assistant Bot.

Appendix A.1.3. Termination Prompt

The prompt used for determining whether a conversation between the user and system has ended or is repeating is as follows:
You will be given a conversation history between a user and an assistant chatbot.
Using it, determine if the conversation has ENDED or is REPEATING between the bot and user.
Your output should be EITHER YES or NO.
An EXAMPLE of how it should be responded is provided below:
user_prompt =
History:
— smith is now talking —
user: hi
bot: Before we begin, please provide your full name.
user: Gabriel Smith
bot: Thank you Gabriel Smith. Welcome! How should I call you?
user: How about Darth Vader
bot: Nice to meet you, Darth Vader. How can I assist you, Darth Vader?
Based on the history above, is the conversation between the user and the bot ENDED or REPEATING?
CORRECT RESPONSE:
NO
An EXAMPLE of how it should be responded is provided below:
user_prompt =
History:
— stacy is now talking —
user: You can call me Gwen.
bot: Nice to meet you, Gwen. What do you need next, Gwen?
user: I need to send a message to Ben about studying together.
bot: Okay! Would you like to send this message anonymously?
user: No, I don’t need to send the message anonymously.
bot: Alright! Can you please tell me the message that you want to send to Ben Smith?
user: Sure! Here’s the message I want to send to Ben: “Hey Ben, I was wondering if you’d be interested in studying together for our Intro to CS class. I find that collaborating with someone helps me understand the material better. Let me know if you’re available and we can set up a study session. Thanks!”
bot: Thank you! I’ll send off your message.
user: Great, thank you! I appreciate your help. I’ll let you know if there’s anything else I need help with. Have a great day!
bot: Of course! It’s my pleasure.
Based on the history above, is the conversation between the user and the bot ENDED or REPEATING?
CORRECT RESPONSE:
YES
An EXAMPLE of how it should be responded is provided below:
user_prompt =
History:
— patel is now talking —
user: You can call me Aisha.
bot: Nice to meet you, Aisha. What can I do for you, Aisha?
user: I would like to make an appointment with Professor Johnson for help with hash tables.
bot: Okay! Please tell me your first choice for a meeting day and time with Amanda Johnson.
user: I would prefer to meet with Professor Johnson on Thursday afternoon around 3:00 PM.
bot: Ok, I think I got it. Is this what you meant?
03/07/2024 (03:00 PM) to 03/07/2024 (03:30 PM)
user: Yes, that’s correct. Thank you for scheduling the appointment with Professor Johnson on Thursday afternoon from 3:00 PM to 3:30 PM.
bot: Awesome, I’ll see if they are free. Furthermore, where do you want to meet with Amanda Johnson?
Based on the history above, is the conversation between the user and the bot ENDED or REPEATING?
CORRECT RESPONSE:
NO
History:
{history}
Based on the history above, is the conversation between the user and the bot ENDED or REPEATING?
Each label corresponds to the following information from the selected user profile:
  • history: Chat history with Assistant Bot.

Appendix A.2. Conversation Examples Demonstrating Assistant Bot Tasks

Figure A1. An example of an initial conversation between the Assistant Bot and a mock user with the name calling function.
Figure A1. An example of an initial conversation between the Assistant Bot and a mock user with the name calling function.
Information 15 00580 g0a1
Figure A2. An example of the Assistant Bot handling an appointment creation request.
Figure A2. An example of the Assistant Bot handling an appointment creation request.
Information 15 00580 g0a2
Figure A3. An example of the Assistant Bot handling a message sending request.
Figure A3. An example of the Assistant Bot handling a message sending request.
Information 15 00580 g0a3
Figure A4. An example of the Assistant Bot handling an inbox check.
Figure A4. An example of the Assistant Bot handling an inbox check.
Information 15 00580 g0a4
Figure A5. An example of the Assistant Bot handling a group creation request.
Figure A5. An example of the Assistant Bot handling a group creation request.
Information 15 00580 g0a5
Figure A6. An example of the Assistant Bot handling a group identification request.
Figure A6. An example of the Assistant Bot handling a group identification request.
Information 15 00580 g0a6
Figure A7. An example of the Assistant Bot sending feedback requests to groups.
Figure A7. An example of the Assistant Bot sending feedback requests to groups.
Information 15 00580 g0a7
Figure A8. An example of the Assistant Bot recording feedback from group members.
Figure A8. An example of the Assistant Bot recording feedback from group members.
Information 15 00580 g0a8
Figure A9. An example of the Assistant Bot handling feedback checks and providing a cohesive summary.
Figure A9. An example of the Assistant Bot handling feedback checks and providing a cohesive summary.
Information 15 00580 g0a9

Appendix A.3. Examples of Assistant Bot and Simulation Fails

Figure A10. An example of the Assistant Bot not handling the check inboxes tasks due to the STDM State Lock. The user’s request for a particular task is ignored by the Assistant Bot due to the chatbot being locked in the name-calling component, thus not allowing it to identify the task requested. The inappropriate response is in red.
Figure A10. An example of the Assistant Bot not handling the check inboxes tasks due to the STDM State Lock. The user’s request for a particular task is ignored by the Assistant Bot due to the chatbot being locked in the name-calling component, thus not allowing it to identify the task requested. The inappropriate response is in red.
Information 15 00580 g0a10
Figure A11. An example of the Assistant Bot not handling the user’s requested task due to STDM NATEX issues. Expressions of gratitude are captured by NATEX groups, but despite having lower priority, the Assistant Bot captured the user’s “Thank you!” rather than the task request through NATEX and generated the inappropriate response. The inappropriate response is in red.
Figure A11. An example of the Assistant Bot not handling the user’s requested task due to STDM NATEX issues. Expressions of gratitude are captured by NATEX groups, but despite having lower priority, the Assistant Bot captured the user’s “Thank you!” rather than the task request through NATEX and generated the inappropriate response. The inappropriate response is in red.
Information 15 00580 g0a11
Figure A12. An example of the Assistant Bot returning to the name calling component due to STDM system bugs. The user’s request for a message included multiple new lines, which the STDM cannot handle, and thus, the Assistant Bot transitioned back to the start to mitigate the issue. The inappropriate response is in red.
Figure A12. An example of the Assistant Bot returning to the name calling component due to STDM system bugs. The user’s request for a message included multiple new lines, which the STDM cannot handle, and thus, the Assistant Bot transitioned back to the start to mitigate the issue. The inappropriate response is in red.
Information 15 00580 g0a12
Figure A13. An example of the Assistant Bot being unable to address multiple tasks due to the STDM multiple task requests error. When the user requested multiple tasks, the Assistant Bot’s response addressed only one of the tasks, as it is only able to move to one component which in turn handles one of the tasks. The inappropriate response is in red.
Figure A13. An example of the Assistant Bot being unable to address multiple tasks due to the STDM multiple task requests error. When the user requested multiple tasks, the Assistant Bot’s response addressed only one of the tasks, as it is only able to move to one component which in turn handles one of the tasks. The inappropriate response is in red.
Information 15 00580 g0a13
Figure A14. An example of the Assistant Bot being unable to extract appointment reason due to the NLU-DST multiple state extraction error. The appointment reason that is mentioned in L1 is not properly extracted due to the STDM framework in which the reason was extracted last. The inappropriate response is in red.
Figure A14. An example of the Assistant Bot being unable to extract appointment reason due to the NLU-DST multiple state extraction error. The appointment reason that is mentioned in L1 is not properly extracted due to the STDM framework in which the reason was extracted last. The inappropriate response is in red.
Information 15 00580 g0a14
Figure A15. An example of the Assistant Bot incorrectly identifying a task due to NLU-IC context misinterpretation. The system misinterprets “checking for new announcements” as “checking for feedback.” The inappropriate response is in red.
Figure A15. An example of the Assistant Bot incorrectly identifying a task due to NLU-IC context misinterpretation. The system misinterprets “checking for new announcements” as “checking for feedback.” The inappropriate response is in red.
Information 15 00580 g0a15
Figure A16. An example of the virtual simulation’s ignoring chatbot issue. The simulated user ignores the Assistant Bot’s last response, instead directly asking for a task to be completed. The inappropriate response is in red.
Figure A16. An example of the virtual simulation’s ignoring chatbot issue. The simulated user ignores the Assistant Bot’s last response, instead directly asking for a task to be completed. The inappropriate response is in red.
Information 15 00580 g0a16
Figure A17. An example of the virtual simulation’s conversation history error. The simulated user had forgotten that the user already requested the Assistant Bot to “call me Diego”, thus repeating the task. The inappropriate response is in red.
Figure A17. An example of the virtual simulation’s conversation history error. The simulated user had forgotten that the user already requested the Assistant Bot to “call me Diego”, thus repeating the task. The inappropriate response is in red.
Information 15 00580 g0a17
Figure A18. An example of the virtual simulation’s user information error. The simulated user, posing as a professor teaching Introduction to Psychology, asks the Assistant Bot for assistance finding a classmate for Data Structures homework, a task that students would request rather than a professor. The inappropriate response is in red. The relevant user information is provided below the conversation.
Figure A18. An example of the virtual simulation’s user information error. The simulated user, posing as a professor teaching Introduction to Psychology, asks the Assistant Bot for assistance finding a classmate for Data Structures homework, a task that students would request rather than a professor. The inappropriate response is in red. The relevant user information is provided below the conversation.
Information 15 00580 g0a18
Figure A19. An example of the virtual simulation’s conversational partner error. The simulation forgets that it is speaking to the Assistant Bot instead of the sender of the appointment request, and thus, the response directly address the sender of the request rather than the Assistant Bot. The inappropriate response is in red.
Figure A19. An example of the virtual simulation’s conversational partner error. The simulation forgets that it is speaking to the Assistant Bot instead of the sender of the appointment request, and thus, the response directly address the sender of the request rather than the Assistant Bot. The inappropriate response is in red.
Information 15 00580 g0a19
Figure A20. An example of the virtual simulation’s role reversal error. The simulation forgets that it is posing as the user and not the Assistant Bot, and thus, the response resembles that of an assistant chatbot rather than the user. The inappropriate response is in red.
Figure A20. An example of the virtual simulation’s role reversal error. The simulation forgets that it is posing as the user and not the Assistant Bot, and thus, the response resembles that of an assistant chatbot rather than the user. The inappropriate response is in red.
Information 15 00580 g0a20

References

  1. Chandan, A.J.; Chattopadhyay, M.; Swaroop Sahoo, S. Implementing Chatbot in Educational Institutes. IJRAR J. 2019, 6, 44–47. [Google Scholar]
  2. Oliveira, P.F.; Matos, P. Introducing a Chatbot to the Web Portal of a Higher Education Institution to Enhance Student Interaction. Eng. Proc. 2023, 56, 128. [Google Scholar] [CrossRef]
  3. Xiao, Z.; Zhou, M.X.; Chen, W.; Yang, H.; Chi, C. If I Hear You Correctly: Building and Evaluating Interview Chatbots with Active Listening Skills. In Proceedings of the CHI Conference on Human Factors in Computing Systems, Honolulu, HI, USA, 25–30 April 2020. [Google Scholar] [CrossRef]
  4. Gebreab, S.A.; Salah, K.; Jayaraman, R.; Habib ur Rehman, M.; Ellaham, S. LLM-Based Framework for Administrative Task Automation in Healthcare. In Proceedings of the 2024 12th International Symposium on Digital Forensics and Security (ISDFS), San Antonio, TX, USA, 29–30 April 2024; pp. 1–7. [Google Scholar] [CrossRef]
  5. Hays, S.; White, D.J. Employing LLMs for Incident Response Planning and Review. arXiv 2024, arXiv:2403.01271. [Google Scholar]
  6. Musumeci, E.; Brienza, M.; Suriani, V.; Nardi, D.; Bloisi, D.D. LLM Based Multi-agent Generation of Semi-structured Documents from Semantic Templates in the Public Administration Domain. In Proceedings of the International Conference on Human–Computer Interaction, Washington, DC, USA, 29 June–4 July 2024; pp. 98–117. [Google Scholar]
  7. Jung, H.; Yeen, H.; Lee, J.; Kim, M.; Bang, N.; Koo, M.W. Enhancing Task-Oriented Dialog System with Subjective Knowledge: A Large Language Model-based Data Augmentation Framework. In Proceedings of the 11th Dialog System Technology Challenge, Prague, Czech Republic, 11 September 2023; Chen, Y.N., Crook, P., Galley, M., Ghazarian, S., Gunasekara, C., Gupta, R., Hedayatnia, B., Kottur, S., Moon, S., Zhang, C., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2023; pp. 150–165. [Google Scholar]
  8. Pan, Y.; Cadamuro, D.; Groh, G. Data-Augmented Task-Oriented Dialogue Response Generation with Domain Adaptation. In Proceedings of the 37th Pacific Asia Conference on Language, Information and Computation, Hong Kong, China, 2–4 December 2023; Huang, C.R., Harada, Y., Kim, J.B., Chen, S., Hsu, Y.Y., Chersoni, E., Pranav, A., Zeng, W.H., Peng, B., Li, Y., et al., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2023; pp. 96–106. [Google Scholar]
  9. Mendoza, S.; Hernández-León, M.; Sánchez-Adame, L.M.; Rodríguez, J.; Decouchant, D.; Meneses-Viveros, A. Supporting Student-Teacher Interaction through a Chatbot. In Proceedings of the Learning and Collaboration Technologies. Human and Technology Ecosystems, Copenhagen, Denmark, 19–24 July 2020; Zaphiris, P., Ioannou, A., Eds.; Springer: Cham, Switzerland, 2020; pp. 93–107. [Google Scholar]
  10. Wang, H.; Wang, L.; Du, Y.; Chen, L.; Zhou, J.; Wang, Y.; Wong, K.F. A Survey of the Evolution of Language Model-Based Dialogue Systems. arXiv 2023, arXiv:2311.16789. [Google Scholar]
  11. Yi, Z.; Ouyang, J.; Liu, Y.; Liao, T.; Xu, Z.; Shen, Y. A Survey on Recent Advances in LLM-Based Multi-turn Dialogue Systems. arXiv 2024, arXiv:2402.18013. [Google Scholar]
  12. Sun, J.; Tian, Y.; Zhou, W.; Xu, N.; Hu, Q.; Gupta, R.; Wieting, J.; Peng, N.; Ma, X. Evaluating Large Language Models on Controlled Generation Tasks. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, Singapore, 6–10 December 2023; Bouamor, H., Pino, J., Bali, K., Eds.; Singapore: Cham, Switzerland, 2023; pp. 3155–3168. [Google Scholar] [CrossRef]
  13. Zhu, Z.; Sun, Z.; Yang, Y. HaluEval-Wild: Evaluating Hallucinations of Language Models in the Wild. arXiv 2024, arXiv:2403.04307. [Google Scholar]
  14. Sekulic, I.; Terragni, S.; Guimarães, V.; Khau, N.; Guedes, B.; Filipavicius, M.; Manso, A.F.; Mathis, R. Reliable LLM-based User Simulator for Task-Oriented Dialogue Systems. In Proceedings of the 1st Workshop on Simulating Conversational Intelligence in Chat (SCI-CHAT 2024); Graham, Y., Liu, Q., Lampouras, G., Iacobacci, I., Madden, S., Khalid, H., Qureshi, R., Eds.; Association for Computational Linguistics: St. Julians, Malta, 2024; pp. 19–35. [Google Scholar]
  15. Wan, D.; Zhang, Z.; Zhu, Q.; Liao, L.; Huang, M. A Unified Dialogue User Simulator for Few-shot Data Augmentation. In Proceedings of the Findings of the Association for Computational Linguistics: EMNLP 2022, Abu Dhabi, United Arab Emirates, 7–11 December 2022; Goldberg, Y., Kozareva, Z., Zhang, Y., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2022; pp. 3788–3799. [Google Scholar] [CrossRef]
  16. Liu, Y.; Jiang, X.; Yin, Y.; Wang, Y.; Mi, F.; Liu, Q.; Wan, X.; Wang, B. One Cannot Stand for Everyone! Leveraging Multiple User Simulators to train Task-oriented Dialogue Systems. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, Toronto, ON, Canada, 9–14 July 2023; Long Papers. Rogers, A., Boyd-Graber, J., Okazaki, N., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2023; Volume 1, pp. 1–21. [Google Scholar] [CrossRef]
  17. Lu, X.; Wang, X. Generative Students: Using LLM-Simulated Student Profiles to Support Question Item Evaluation. In Proceedings of the 11th ACM Conference on Learning @ Scale, New York, NY, USA, 18–20 July 2024; pp. 16–27. [Google Scholar] [CrossRef]
  18. Touvron, H.; Martin, L.; Stone, K.; Albert, P.; Almahairi, A.; Babaei, Y.; Bashlykov, N.; Batra, S.; Bhargava, P.; Bhosale, S.; et al. Llama 2: Open Foundation and Fine-Tuned Chat Models. arXiv 2023, arXiv:2307.09288. [Google Scholar]
  19. Jiang, A.Q.; Sablayrolles, A.; Mensch, A.; Bamford, C.; Chaplot, D.S.; de las Casas, D.; Bressand, F.; Lengyel, G.; Lample, G.; Saulnier, L.; et al. Mistral 7B. arXiv 2023, arXiv:2310.06825. [Google Scholar]
  20. OpenAI; Achiam, J.; Adler, S.; Agarwal, S.; Ahmad, L.; Akkaya, I.; Aleman, F.L.; Almeida, D.; Altenschmidt, J.; Altman, S.; et al. GPT-4 Technical Report. arXiv 2024, arXiv:2303.08774. [Google Scholar]
  21. Zhang, J.; Bui, T.; Yoon, S.; Chen, X.; Liu, Z.; Xia, C.; Tran, Q.H.; Chang, W.; Yu, P. Few-Shot Intent Detection via Contrastive Pre-Training and Fine-Tuning. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, Punta Cana, Dominican Republic, 7–11 November 2021; Moens, M.F., Huang, X., Specia, L., Yih, S.W.t., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2021; pp. 1906–1912. [Google Scholar] [CrossRef]
  22. Parikh, S.; Tiwari, M.; Tumbade, P.; Vohra, Q. Exploring Zero and Few-shot Techniques for Intent Classification. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, Toronto, ON, Canada, 9–14 July 2024; Industry Track; Sitaram, S., Beigman Klebanov, B., Williams, J.D., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2023; pp. 744–751. [Google Scholar] [CrossRef]
  23. Heck, M.; Lubis, N.; Ruppik, B.; Vukovic, R.; Feng, S.; Geishauser, C.; Lin, H.C.; van Niekerk, C.; Gasic, M. ChatGPT for Zero-shot Dialogue State Tracking: A Solution or an Opportunity? In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, Toronto, ON, Canada, 9–14 July 2024; Volume 2: Short Papers; Rogers, A., Boyd-Graber, J., Okazaki, N., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2023; Volume 2, pp. 936–950. [Google Scholar] [CrossRef]
  24. Zang, X.; Rastogi, A.; Sunkara, S.; Gupta, R.; Zhang, J.; Chen, J. MultiWOZ 2.2: A Dialogue Dataset with Additional Annotation Corrections and State Tracking Baselines. In Proceedings of the 2nd Workshop on Natural Language Processing for Conversational AI, Online, 9 July 2020; Wen, T.H., Celikyilmaz, A., Yu, Z., Papangelis, A., Eric, M., Kumar, A., Casanueva, I., Shah, R., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2020; pp. 109–117. [Google Scholar] [CrossRef]
  25. Rastogi, A.; Zang, X.; Sunkara, S.; Gupta, R.; Khaitan, P. Towards Scalable Multi-Domain Conversational Agents: The Schema-Guided Dialogue Dataset. In Proceedings of the AAAI Conference on Artificial Intelligence, New York, NY, USA, 7–12 February 2020; pp. 8689–8696. [Google Scholar]
  26. Feng, Y.; Lu, Z.; Liu, B.; Zhan, L.; Wu, X.M. Towards LLM-driven Dialogue State Tracking. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, Singapore, 6–10 December 2023; Bouamor, H., Pino, J., Bali, K., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2023; pp. 739–755. [Google Scholar] [CrossRef]
  27. Bang, Y.; Cahyawijaya, S.; Lee, N.; Dai, W.; Su, D.; Wilie, B.; Lovenia, H.; Ji, Z.; Yu, T.; Chung, W.; et al. A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity. In Proceedings of the 13th International Joint Conference on Natural Language Processing and the 3rd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics, Bali, Indonesia, 1–4 November 2023; Long Papers. Park, J.D.C., Arase, Y., Hu, B., Lu, W., Wijaya, D., Purwarianti, A., Krisnadhi, A.A., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2023; Volume 1, pp. 675–718. [Google Scholar] [CrossRef]
  28. Chung, W.; Cahyawijaya, S.; Wilie, B.; Lovenia, H.; Fung, P. InstructTODS: Large Language Models for End-to-End Task-Oriented Dialogue Systems. In Proceedings of the 2nd Workshop on Natural Language Interfaces, Bali, Indonesia, 1 November 2023; Chen, K., Ku, L.W., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2023; pp. 1–21. [Google Scholar] [CrossRef]
  29. Hudeček, V.; Dusek, O. Are Large Language Models All You Need for Task-Oriented Dialogue? In Proceedings of the 24th Annual Meeting of the Special Interest Group on Discourse and Dialogue, Prague, Czech Republic, 11–15 September 2023; Stoyanchev, S., Joty, S., Schlangen, D., Dusek, O., Kennington, C., Alikhani, M., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2023; pp. 216–228. [Google Scholar] [CrossRef]
  30. Toxtli, C.; Monroy-Hernández, A.; Cranshaw, J. Understanding Chatbot-mediated Task Management. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems, New York, NY, USA, 21–26 April 2018; pp. 1–6. [Google Scholar] [CrossRef]
  31. Deriu, J.M.; Cieliebak, M. Towards a Metric for Automated Conversational Dialogue System Evaluation and Improvement. In Proceedings of the 12th International Conference on Natural Language Generation, Tokyo, Japan, 29 October–1 November 2019; van Deemter, K., Lin, C., Takamura, H., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2019; pp. 432–437. [Google Scholar] [CrossRef]
  32. Ghandeharioun, A.; Shen, J.H.; Jaques, N.; Ferguson, C.; Jones, N.; Lapedriza, A.; Picard, R. Approximating Interactive Human Evaluation with Self-Play for Open-Domain Dialog Systems. In Proceedings of the Advances in Neural Information Processing Systems, Vancouver, QC, Canada, 8–14 December 2019; Wallach, H., Larochelle, H., Beygelzimer, A., d’Alché-Buc, F., Fox, E., Garnett, R., Eds.; Curran Associates, Inc.: Red Hook, NY, USA, 2019; Volume 32. [Google Scholar]
  33. Deriu, J.; Tuggener, D.; von Däniken, P.; Campos, J.A.; Rodrigo, A.; Belkacem, T.; Soroa, A.; Agirre, E.; Cieliebak, M. Spot The Bot: A Robust and Efficient Framework for the Evaluation of Conversational Dialogue Systems. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), Online, 16–20 November 2020; Webber, B., Cohn, T., He, Y., Liu, Y., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2020; pp. 3971–3984. [Google Scholar] [CrossRef]
  34. Yang, R.; Li, Z.; Tang, H.; Zhu, K. ChatMatch: Evaluating Chatbots by Autonomous Chat Tournaments. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics, Dublin, Ireland, 22–27 May 2022; Muresan, S., Nakov, P., Villavicencio, A., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2022; Volume 1, pp. 7579–7590. [Google Scholar] [CrossRef]
  35. Abeysinghe, B.; Circi, R. The Challenges of Evaluating LLM Applications: An Analysis of Automated, Human, and LLM-Based Approaches. arXiv 2024, arXiv:2406.03339. [Google Scholar]
  36. Chiang, C.H.; Lee, H.y. Can Large Language Models Be an Alternative to Human Evaluations? In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, Toronto, ON, Canada, 9–14 July 202; Long Papers; Rogers, A., Boyd-Graber, J., Okazaki, N., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2023; Volume 1, pp. 15607–15631. [Google Scholar] [CrossRef]
  37. Svikhnushina, E.; Pu, P. Approximating Online Human Evaluation of Social Chatbots with Prompting. In Proceedings of the 24th Annual Meeting of the Special Interest Group on Discourse and Dialogue, Prague, Czech Republic, 11–15 September 2023; Stoyanchev, S., Joty, S., Schlangen, D., Dusek, O., Kennington, C., Alikhani, M., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2023; pp. 268–281. [Google Scholar] [CrossRef]
  38. Park, J.S.; O’Brien, J.D.C.; Cai, C.J.; Morris, M.R.; Liang, P.; Bernstein, M.S. Generative Agents: Interactive Simulacra of Human Behavior. arXiv 2023, arXiv:2304.03442. [Google Scholar]
  39. Zhou, X.; Zhu, H.; Mathur, L.; Zhang, R.; Yu, H.; Qi, Z.; Morency, L.P.; Bisk, Y.; Fried, D.; Neubig, G.; et al. SOTOPIA: Interactive Evaluation for Social Intelligence in Language Agents. arXiv 2024, arXiv:2310.11667. [Google Scholar]
  40. Finch, J.D.; Choi, J.D. Emora STDM: A Versatile Framework for Innovative Dialogue System Development. In Proceedings of the 21th Annual Meeting of the Special Interest Group on Discourse and Dialogue, Virtual Meeting, 1–3 July 2020; Pietquin, O., Muresan, S., Chen, V., Kennington, C., Vandyke, D., Dethlefs, N., Inoue, K., Ekstedt, E., Ultes, S., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2023; pp. 261–264. [Google Scholar] [CrossRef]
  41. Madotto, A.; Liu, Z.; Lin, Z.; Fung, P. Language Models as Few-Shot Learner for Task-Oriented Dialogue Systems. arXiv 2020, arXiv:2008.06239. [Google Scholar]
  42. Brown, T.; Mann, B.; Ryder, N.; Subbiah, M.; Kaplan, J.D.; Dhariwal, P.; Neelakantan, A.; Shyam, P.; Sastry, G.; Askell, A.; et al. Language Models are Few-Shot Learners. In Proceedings of the Advances in Neural Information Processing Systems, Online, 6–12 December 2020; Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M., Lin, H., Eds.; Curran Associates, Inc.: Red Hook, NY, USA, 2020; Volume 33, pp. 1877–1901. [Google Scholar]
  43. Yuan, C.; Xie, Q.; Ananiadou, S. Zero-shot Temporal Relation Extraction with ChatGPT. In Proceedings of the 22nd Workshop on Biomedical Natural Language Processing and BioNLP Shared Tasks, Toronto, ON, Canada, 13 July 2023; Demner-Fushman, D., Ananiadou, S., Cohen, K., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2023; pp. 92–102. [Google Scholar] [CrossRef]
  44. Wei, X.; Cui, X.; Cheng, N.; Wang, X.; Zhang, X.; Huang, S.; Xie, P.; Xu, J.; Chen, Y.; Zhang, M.; et al. Zero-Shot Information Extraction via Chatting with ChatGPT. arXiv 2023, arXiv:2302.10205. [Google Scholar]
  45. Salewski, L.; Alaniz, S.; Rio-Torto, I.; Schulz, E.; Akata, Z. In-Context Impersonation Reveals Large Language Models’ Strengths and Biases. In Proceedings of the 37th Conference on Neural Information Processing Systems, New Orleans, LA, USA, 10–16 December 2023. [Google Scholar]
  46. Kim, H.; Hessel, J.; Jiang, L.; West, P.; Lu, X.; Yu, Y.; Zhou, P.; Bras, R.; Alikhani, M.; Kim, G.; et al. SODA: Million-scale Dialogue Distillation with Social Commonsense Contextualization. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, Singapore, 6–10 December 2023; Bouamor, H., Pino, J., Bali, K., Eds.; Association for Computational Linguistics: Kerrville, TX, USA, 2023; pp. 12930–12949. [Google Scholar] [CrossRef]
Figure 1. A diagram of the Assistant Bot’s system framework and execution process.
Figure 1. A diagram of the Assistant Bot’s system framework and execution process.
Information 15 00580 g001
Figure 2. An example of synchronous multi-party dialogue among two users, Gwen and Peter, scheduling a meeting assisted by our chatbot. Initially, the chatbot engages with Gwen, receiving her request intended for Peter (L1-9). Subsequently, it interacts with Peter to ascertain availability and receives a request for rescheduling (L10-16). Finally, the chatbot reconnects with Gwen to confirm the rescheduled appointment from Peter (L17-21).
Figure 2. An example of synchronous multi-party dialogue among two users, Gwen and Peter, scheduling a meeting assisted by our chatbot. Initially, the chatbot engages with Gwen, receiving her request intended for Peter (L1-9). Subsequently, it interacts with Peter to ascertain availability and receives a request for rescheduling (L10-16). Finally, the chatbot reconnects with Gwen to confirm the rescheduled appointment from Peter (L17-21).
Information 15 00580 g002
Figure 3. A dialogue example demonstrating how our chatbot utilizes GPT for NLU.
Figure 3. A dialogue example demonstrating how our chatbot utilizes GPT for NLU.
Information 15 00580 g003
Figure 4. The dashboard (left) and chat interface (right) of our mobile app hosting the Assistant Bot.
Figure 4. The dashboard (left) and chat interface (right) of our mobile app hosting the Assistant Bot.
Information 15 00580 g004
Table 1. Description of tasks that Assistant Bot can execute.
Table 1. Description of tasks that Assistant Bot can execute.
TypeTaskDescriptionFigures
N&HSet nameEstablishes the names that the user wishes to be called byFigure A1
List rosterLists the users who are registered in the database
APPMake appointmentSends an appointment request to a different userFigure A2
Confirm appointmentConfirms an appointment request between two users and adds to schedules
Edit appointmentEdits an appointment request’s details
Reschedule appointmentReschedules appointment request to different date
Cancel appointmentCancels appointment request
MSGSend messageSends message to recipientFigure A3
Check inboxChecks inbox for messages and appointment requestsFigure A4
GRPMake groupMakes group of multiple peopleFigure A5
Delete groupDeletes a group that has been created (must be leader of the group to proceed)
Leave groupLeaves a group that user has been added to (must be a group member to proceed)
Check groupsChecks groups that users are inFigure A6
Send announcementSends an identical message to group members (must be leader of the group to proceed)
Request feedbackSends feedback questions to group members (must be leader of the group to proceed)Figure A7
Send feedbackRecords responses to feedback question (must be a group member to proceed)Figure A8
Check feedbackChecks statistics of feedback question (must be leader of the group to proceed)Figure A9
Table 2. Success and error rates for task completion, natural language understanding, and the virtual simulator across the 10 evaluation metrics (Section 5.3). Vir-Avg: the average rates across the five scenarios (Section 5.1). Hum-Avg: the average rates within the human simulation (Section 5.2). Bolded values represent the highest success rates.
Table 2. Success and error rates for task completion, natural language understanding, and the virtual simulator across the 10 evaluation metrics (Section 5.3). Vir-Avg: the average rates across the five scenarios (Section 5.1). Hum-Avg: the average rates within the human simulation (Section 5.2). Bolded values represent the highest success rates.
ScenarioTask CompletionNatural Language UnderstandingSimulator
( u , g ) NLUSTDSUCSWSINTDSTSTDSUCSWSAPP
(10, 1)9.336.754.085.37.11.822.468.788.769.8
(20, 1)7.834.657.687.84.63.023.968.590.081.8
(20, 2)6.838.354.988.74.52.922.570.190.479.6
(20, 3)6.237.756.190.03.65.320.970.288.783.6
(30, 1)9.741.049.383.55.54.620.469.487.277.0
Vir-Avg8.037.654.487.15.13.522.069.489.080.3
Hum-Avg10.543.146.481.65.24.226.064.687.3-
Table 3. Analysis of errors in the Assistant Bot. STDM: errors caused by the STDM limitations; NLU-DST/IC: NLU errors caused by GPT’s misinterpretation of dialogue state tracking & intent classification, respectively.
Table 3. Analysis of errors in the Assistant Bot. STDM: errors caused by the STDM limitations; NLU-DST/IC: NLU errors caused by GPT’s misinterpretation of dialogue state tracking & intent classification, respectively.
CategoryError Type%
STDMState Lock0.68
NATEX Issues0.20
System Bugs0.11
Multiple Task Requests0.01
NLU-DSTMultiple State Extraction0.84
Information Mis-extraction0.16
NLU-ICContext Mis-interpretation0.62
Task Mis-classification0.38
Table 4. Error analysis on the virtual simulation.
Table 4. Error analysis on the virtual simulation.
Error Type%
Ignoring Context0.39
Forgetting Conversation History0.30
Forgetting User Information0.19
Confusing Conversational Partner0.06
Role Reverse0.06
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Paek, E.S.; Fan, T.; Finch, J.D.; Choi, J.D. Enhancing Task-Oriented Dialogue Systems through Synchronous Multi-Party Interaction and Multi-Group Virtual Simulation. Information 2024, 15, 580. https://doi.org/10.3390/info15090580

AMA Style

Paek ES, Fan T, Finch JD, Choi JD. Enhancing Task-Oriented Dialogue Systems through Synchronous Multi-Party Interaction and Multi-Group Virtual Simulation. Information. 2024; 15(9):580. https://doi.org/10.3390/info15090580

Chicago/Turabian Style

Paek, Ellie S., Talyn Fan, James D. Finch, and Jinho D. Choi. 2024. "Enhancing Task-Oriented Dialogue Systems through Synchronous Multi-Party Interaction and Multi-Group Virtual Simulation" Information 15, no. 9: 580. https://doi.org/10.3390/info15090580

APA Style

Paek, E. S., Fan, T., Finch, J. D., & Choi, J. D. (2024). Enhancing Task-Oriented Dialogue Systems through Synchronous Multi-Party Interaction and Multi-Group Virtual Simulation. Information, 15(9), 580. https://doi.org/10.3390/info15090580

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop