Jie Tao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Jie Tao has a total of 28 co-authors. Here are the top co-authors by number of shared publications:
- Lina Zhou
University of North Carolina at Charlotte, Charlotte, USA
9 shared publications - Xing Fang
School of Information Technology, Illinois State University, Normal, IL 61761, USA
6 shared publications - Amit V. Deokar
Manning School of Business, University of Massachusetts Lowell, Lowell, MA, USA
3 shared publications - Lina Zhou
University of North Carolina at Charlotte, Charlotte, USA
3 shared publications Conor Feeney
Fairfield University Fairfield 06824 CT USA
1 shared publication- Dongsong Zhang
Belk College of Business & The School of Data Science, The University of North Carolina at Charlotte, Charlotte, North Carolina 28223
1 shared publication
Latest Publications
- Open Access
- 5 September 2026
- Published by Springer Nature in Information Systems Frontiers
- p. 1-26
From Explainability to Actionability: a Tiered Adaptable Multi-Agent Framework with Agent Reasoning Tools for Collaborative Failure Recovery
Effective human-AI collaboration, especially in failure scenarios, requires systems that function as active partners rather than static tools. This research addresses this requirement by introducing a tiered multi-agent architecture grounded in Human-Centered eXplainable AI principles. The architecture consists of three novel design artifacts: tiered reasoning that adapts explanation depth to failure severity, a traceable memory bus for auditability, and flexible reasoning tools for enhanced adaptability. These designs enable agents to not only perform evidence-based diagnoses of performance gaps but devise recovery strategies and propose actionable improvement plans as well. We empirically evaluate this architecture on an aspect term extraction task using hybrid methods that combine performance comparisons against state-of-the-art baselines, human expert user studies, and multi-role user simulations. The results demonstrate that our architecture significantly enhances both task performance and failure recovery diagnosis and plans. We then validate the generalizability of our architecture with a second task of comparable complexity. This research contributes an adaptable and generalizable framework and foundational artifacts for trustworthy AI teammates.
- 5 September 2026
- Published by Springer Nature in Information Systems Frontiers
- p. 1-26
Open AccessFrom Explainability to Actionability: a Tiered Adaptable Multi-Agent Framework with Agent Reasoning Tools for Collaborative Failure Recovery
Effective human-AI collaboration, especially in failure scenarios, requires systems that function as active partners rather than static tools. This research addresses this requirement by introducing a tiered multi-agent architecture grounded in Human-Centered eXplainable AI principles. The architecture consists of three novel design artifacts: tiered reasoning that adapts explanation depth to failure severity, a traceable memory bus for auditability, and flexible reasoning tools for enhanced adaptability. These designs enable agents to not only perform evidence-based diagnoses of performance gaps but devise recovery strategies and propose actionable improvement plans as well. We empirically evaluate this architecture on an aspect term extraction task using hybrid methods that combine performance comparisons against state-of-the-art baselines, human expert user studies, and multi-role user simulations. The results demonstrate that our architecture significantly enhances both task performance and failure recovery diagnosis and plans. We then validate the generalizability of our architecture with a second task of comparable complexity. This research contributes an adaptable and generalizable framework and foundational artifacts for trustworthy AI teammates.
- Open Access
- 23 March 2026
- Published by Springer Nature in Information Systems Frontiers
- Vol. 28 (3) , 1-17
Toward a Knowledge Discovery Method to Fake Review Detection
Fake reviews can mislead consumers into making wrong or suboptimal decisions. Despite the tremendous effort in advancing fake review detection models, existing work has focused on developing complex models to improve the detection performance (e.g., accuracy); however, there has been little research on discovering knowledge about fake or authentic reviews that can guide the detection efforts from those complex models. In addition, previous detection models treat online reviews as the unit of analysis while ignoring the semantic coherence of different sentences within the same review. Using a combination of deep learning with transformers and unsupervised learning, we propose a model-agnostic and data-independent knowledge discovery method for fake review detection that consists of two components: Domain Adaptation Transfer Learning (DATL) and Fine-Grained Knowledge Discovery (FGKD). DATL encodes review texts by adapting a transformer-based model to the context of online reviews. Drawing on the coherence principle from a deception theory, FGKD first identifies review sentences containing tacit knowledge and then uses them to discover knowledge in support of fake review detection. We further develop novel evaluation metrics beyond accuracy for the generated knowledge by extending those from interpretable machine learning. The empirical evaluation results demonstrate consistently positive effects of our knowledge discovery method on the efficiency of fake review detection without hurting model effectiveness. This study not only contributes to interpretable machine learning but also lays the foundation for AI model design to augment humans’ ability in fake review detection.
- 23 March 2026
- Published by Springer Nature in Information Systems Frontiers
- Vol. 28 (3) , 1-17
Open AccessToward a Knowledge Discovery Method to Fake Review Detection
Fake reviews can mislead consumers into making wrong or suboptimal decisions. Despite the tremendous effort in advancing fake review detection models, existing work has focused on developing complex models to improve the detection performance (e.g., accuracy); however, there has been little research on discovering knowledge about fake or authentic reviews that can guide the detection efforts from those complex models. In addition, previous detection models treat online reviews as the unit of analysis while ignoring the semantic coherence of different sentences within the same review. Using a combination of deep learning with transformers and unsupervised learning, we propose a model-agnostic and data-independent knowledge discovery method for fake review detection that consists of two components: Domain Adaptation Transfer Learning (DATL) and Fine-Grained Knowledge Discovery (FGKD). DATL encodes review texts by adapting a transformer-based model to the context of online reviews. Drawing on the coherence principle from a deception theory, FGKD first identifies review sentences containing tacit knowledge and then uses them to discover knowledge in support of fake review detection. We further develop novel evaluation metrics beyond accuracy for the generated knowledge by extending those from interpretable machine learning. The empirical evaluation results demonstrate consistently positive effects of our knowledge discovery method on the efficiency of fake review detection without hurting model effectiveness. This study not only contributes to interpretable machine learning but also lays the foundation for AI model design to augment humans’ ability in fake review detection.
- Open Access
- 1 January 2025
- Published by HICSS Conference Office
Introduction to the Minitrack on Technological Advancements in Digital Collaboration with Generative AI and Large Language Models
This mini-track explores the evolving landscape of digital collaboration, focusing on the integration of generative AI and large language models (LLMs). Technological advancements, particularly in generative AI and LLMs, are transforming digital collaboration. These technologies offer opportunities to enhance productivity and creativity in various tasks. For instance, LLMs can assist users with open-ended tasks in digital collaborative environments. However, several challenges need to be addressed. One key challenge is the need for more sophisticated and explainable methodologies that can facilitate more nuanced and effective human-AI collaboration. Additionally, ensuring the accuracy and social responsibility of AIgenerated content, particularly in mitigating hallucinations and biases, remains crucial. This minitrack addresses the above challenges by focusing on the advancement of human interaction with Generative AI and LLMs. It aims to explore various topics, including prompt engineering, Parameter Efficient Fine Tuning (PEFT), Retriever Augmented Generation (RAG), and other relevant techniques. The goal is to enhance the quality, productivity, creativity, user satisfaction, and cost efficiency of digital collaboration. This mini-track consists of three papers. They provide a broad array of topics and methods that explore the technological advancements in human collaboration with generative AI and LLM, including: • How can we develop more sophisticated and explainable AI models that facilitate effective human-AI collaboration, moving beyond the limitations of current chatbot interactions? • How can we ensure the accuracy and social responsibility of AI-generated content, particularly in mitigating hallucinations and biases? • What are the effective strategies for prompt engineering, PEFT, RAG, and other techniques to improve the performance and usability of generative AI and LLMs in digital collaboration? • How can we address ethical considerations, including privacy and security concerns, associated with human-AI collaboration in digital environments?
- 1 January 2025
- Published by HICSS Conference Office
Open AccessIntroduction to the Minitrack on Technological Advancements in Digital Collaboration with Generative AI and Large Language Models
This mini-track explores the evolving landscape of digital collaboration, focusing on the integration of generative AI and large language models (LLMs). Technological advancements, particularly in generative AI and LLMs, are transforming digital collaboration. These technologies offer opportunities to enhance productivity and creativity in various tasks. For instance, LLMs can assist users with open-ended tasks in digital collaborative environments. However, several challenges need to be addressed. One key challenge is the need for more sophisticated and explainable methodologies that can facilitate more nuanced and effective human-AI collaboration. Additionally, ensuring the accuracy and social responsibility of AIgenerated content, particularly in mitigating hallucinations and biases, remains crucial. This minitrack addresses the above challenges by focusing on the advancement of human interaction with Generative AI and LLMs. It aims to explore various topics, including prompt engineering, Parameter Efficient Fine Tuning (PEFT), Retriever Augmented Generation (RAG), and other relevant techniques. The goal is to enhance the quality, productivity, creativity, user satisfaction, and cost efficiency of digital collaboration. This mini-track consists of three papers. They provide a broad array of topics and methods that explore the technological advancements in human collaboration with generative AI and LLM, including: • How can we develop more sophisticated and explainable AI models that facilitate effective human-AI collaboration, moving beyond the limitations of current chatbot interactions? • How can we ensure the accuracy and social responsibility of AI-generated content, particularly in mitigating hallucinations and biases? • What are the effective strategies for prompt engineering, PEFT, RAG, and other techniques to improve the performance and usability of generative AI and LLMs in digital collaboration? • How can we address ethical considerations, including privacy and security concerns, associated with human-AI collaboration in digital environments?
- Open Access
- 1 January 2025
- Published by HICSS Conference Office
Is the Larger the Better? An Exploratory Study into Human-Large Language Model Collaboration
Large language models (LLMs) have garnered considerable attention in both academics and industry. Given an array of LLMs available, one of the primary challenges lies in their selection and adaptation strategies. Although LLMs are generally large, they still vary significantly in size and larger LLMs consume significantly more computing resources. This prompts the inquiry into whether larger models perform better. In addition, there is a widespread recognition of the power of LLMs in performing open-ended or generative tasks. However, how to use LLMs to address a closed-ended problem remains under explored. The exploration of human-LLM collaboration on close-ended problem has been even more sparse. This research aims to address the above limitations by comparing different types of state-of-the-art adaptation strategies for LLMs, including in-context learning and fine-tuning. Moreover, it employs multi-class multi-label classification - a close-ended problem to empirically evaluate those adaptation strategies. The research findings provide valuable insights and recommendations for human users considering deploying LLMs for close-ended problems.
- 1 January 2025
- Published by HICSS Conference Office
Open AccessIs the Larger the Better? An Exploratory Study into Human-Large Language Model Collaboration
Large language models (LLMs) have garnered considerable attention in both academics and industry. Given an array of LLMs available, one of the primary challenges lies in their selection and adaptation strategies. Although LLMs are generally large, they still vary significantly in size and larger LLMs consume significantly more computing resources. This prompts the inquiry into whether larger models perform better. In addition, there is a widespread recognition of the power of LLMs in performing open-ended or generative tasks. However, how to use LLMs to address a closed-ended problem remains under explored. The exploration of human-LLM collaboration on close-ended problem has been even more sparse. This research aims to address the above limitations by comparing different types of state-of-the-art adaptation strategies for LLMs, including in-context learning and fine-tuning. Moreover, it employs multi-class multi-label classification - a close-ended problem to empirically evaluate those adaptation strategies. The research findings provide valuable insights and recommendations for human users considering deploying LLMs for close-ended problems.
- Open Access
- 30 September 2024
- Published by Association for Information Systems in AIS Transactions on Human-Computer Interaction
- Vol. 16 (3) , 364-387
Generative AI for Intelligence Augmentation: Categorization and Evaluation Frameworks for Large Language Model Adaptation
Generative AI (GenAI) has transformed how businesses operate and innovate and how individuals learn, live, and work. Large language models (LLMs), a specific type of GenAI, focus on generating human-like text based on user instructions. Like other types of GenAI, LLMs have received wide recognition for their potential to augment human intelligence, but several challenges hinder efforts to realize their full potential in practice. Some notable challenges include not adequately exploring LLM applications beyond chatbots and/or text generation, the difficulty in categorizing various LLM adaptation strategies (particularly regarding human interactions), and the lack of a reference framework for evaluating and selecting LLM adaptation strategies from a human-centered perspective. To address these challenges, we propose a categorization framework for LLM adaptation that features two human-centered dimensions and stage LLM adaptation with respect to when it interacts with human intelligence. Additionally, we introduce an evaluation framework that incorporates a human-centered perspective that goes beyond the common machine-centered measures. Our empirical investigations, in which we use text classification as use cases, not only demonstrate the application of these frameworks but also compare various adaptation strategies. These artifacts and findings provide fresh insights and practical recommendations for selecting effective adaptation strategies to improve the efficacy of LLMs for intelligence augmentation. We further identify future research issues to address current limitations and suggest improvements for the proposed frameworks.
- 30 September 2024
- Published by Association for Information Systems in AIS Transactions on Human-Computer Interaction
- Vol. 16 (3) , 364-387
Open AccessGenerative AI for Intelligence Augmentation: Categorization and Evaluation Frameworks for Large Language Model Adaptation
Generative AI (GenAI) has transformed how businesses operate and innovate and how individuals learn, live, and work. Large language models (LLMs), a specific type of GenAI, focus on generating human-like text based on user instructions. Like other types of GenAI, LLMs have received wide recognition for their potential to augment human intelligence, but several challenges hinder efforts to realize their full potential in practice. Some notable challenges include not adequately exploring LLM applications beyond chatbots and/or text generation, the difficulty in categorizing various LLM adaptation strategies (particularly regarding human interactions), and the lack of a reference framework for evaluating and selecting LLM adaptation strategies from a human-centered perspective. To address these challenges, we propose a categorization framework for LLM adaptation that features two human-centered dimensions and stage LLM adaptation with respect to when it interacts with human intelligence. Additionally, we introduce an evaluation framework that incorporates a human-centered perspective that goes beyond the common machine-centered measures. Our empirical investigations, in which we use text classification as use cases, not only demonstrate the application of these frameworks but also compare various adaptation strategies. These artifacts and findings provide fresh insights and practical recommendations for selecting effective adaptation strategies to improve the efficacy of LLMs for intelligence augmentation. We further identify future research issues to address current limitations and suggest improvements for the proposed frameworks.
- Open Access
- 1 January 2024
- Published by HICSS Conference Office
Beyond the Social Media Contents: The Role of Social Interactions in Stance Detection
Stance detection categorizes the stance toward a specific target or topic, or other aspects of interest (e.g., a social media comment) as favor, against, or neutral. The discourse of stance detection has evolved from court debate to social media, particularly in the analyses of political, social, and health issues. Despite its long-standing history, stance detection still faces significant challenges partly due to ambiguous, diverse, and informal expressions of human language in social media. Motivated by the affordance of social interactions on social media platforms, this study aims to investigate whether social interactions are useful and how to represent and incorporate them into stance detection models effectively. To this end, we propose a framework that integrates graph learning with transformers. The empirical evaluation results with an extended benchmark dataset in the political discourse demonstrate the superior performance of the framework to the state-of-the-art baseline models and highlight the significant role of social interaction networks in stance detection. The framework can also be used to guide the efforts in social media monitoring, marketing, and informed decision-making.
- 1 January 2024
- Published by HICSS Conference Office
Open AccessBeyond the Social Media Contents: The Role of Social Interactions in Stance Detection
Stance detection categorizes the stance toward a specific target or topic, or other aspects of interest (e.g., a social media comment) as favor, against, or neutral. The discourse of stance detection has evolved from court debate to social media, particularly in the analyses of political, social, and health issues. Despite its long-standing history, stance detection still faces significant challenges partly due to ambiguous, diverse, and informal expressions of human language in social media. Motivated by the affordance of social interactions on social media platforms, this study aims to investigate whether social interactions are useful and how to represent and incorporate them into stance detection models effectively. To this end, we propose a framework that integrates graph learning with transformers. The empirical evaluation results with an extended benchmark dataset in the political discourse demonstrate the superior performance of the framework to the state-of-the-art baseline models and highlight the significant role of social interaction networks in stance detection. The framework can also be used to guide the efforts in social media monitoring, marketing, and informed decision-making.
- Open Access
- 5 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 234-240
Toward an Explainable Artificial Intelligence Approach to Enhance Medical Imaging Classification Models
Recent studies in eXplainable Artificial Intelligence (XAI) have been focusing on demystifying neural networks, which are considered algorithmic black-boxes. This is particularly important given the plethora of AI solutions, and in certain domains, such as healthcare and finance, due to regulation and compliance requirements. These XAI approaches yielded discussion points (e.g., evidence of model sufficiency), but overlooked connecting the explanations with steps to enhance model performances, as well as acquiring domain knowledge from them. The key difference is that the data enhancement strategies can improve model performance without re-training. In this study, we propose a novel XAI approach that utilizes visual explanations, in combination with domain knowledge, to guide data enhancement practices. The approach provides guidelines for future data enhancement, as well as knowledge elements in the form of explanation on the enhancement. We evaluated the proposed approach using a medical imaging dataset, and demonstrated the feasibility and effectiveness of the proposed approach.
- 5 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 234-240
Open AccessToward an Explainable Artificial Intelligence Approach to Enhance Medical Imaging Classification Models
Recent studies in eXplainable Artificial Intelligence (XAI) have been focusing on demystifying neural networks, which are considered algorithmic black-boxes. This is particularly important given the plethora of AI solutions, and in certain domains, such as healthcare and finance, due to regulation and compliance requirements. These XAI approaches yielded discussion points (e.g., evidence of model sufficiency), but overlooked connecting the explanations with steps to enhance model performances, as well as acquiring domain knowledge from them. The key difference is that the data enhancement strategies can improve model performance without re-training. In this study, we propose a novel XAI approach that utilizes visual explanations, in combination with domain knowledge, to guide data enhancement practices. The approach provides guidelines for future data enhancement, as well as knowledge elements in the form of explanation on the enhancement. We evaluated the proposed approach using a medical imaging dataset, and demonstrated the feasibility and effectiveness of the proposed approach.
- Open Access
- 1 January 2023
- Published by HICSS Conference Office
On Left and Right: Understanding the Discourse of Presidential Election in Social Media Communities
As a promising platform for political discourse, social media becomes a battleground for presidential candidates as well as their supporters and opponents. Stance detection is one of the key tasks in the understanding of political discourse. However, existing methods are dominated by supervised techniques, which require labeled data. Previous work on stance detection is largely conducted at the post or user level. Despite that some studies have considered online political communities, they either only select a few communities or assume the stance coherence of these communities. Political party extraction has rarely been addressed explicitly. To address the limitations, we developed an unsupervised learning approach to political party extraction and stance detection from social media discourse. We also analyzed and compared (sub)communities with respect to their characteristics of political stances and parties. We further explored (sub)communities’ shift in political stance after the 2020 US presidential election.
- 1 January 2023
- Published by HICSS Conference Office
Open AccessOn Left and Right: Understanding the Discourse of Presidential Election in Social Media Communities
As a promising platform for political discourse, social media becomes a battleground for presidential candidates as well as their supporters and opponents. Stance detection is one of the key tasks in the understanding of political discourse. However, existing methods are dominated by supervised techniques, which require labeled data. Previous work on stance detection is largely conducted at the post or user level. Despite that some studies have considered online political communities, they either only select a few communities or assume the stance coherence of these communities. Political party extraction has rarely been addressed explicitly. To address the limitations, we developed an unsupervised learning approach to political party extraction and stance detection from social media discourse. We also analyzed and compared (sub)communities with respect to their characteristics of political stances and parties. We further explored (sub)communities’ shift in political stance after the 2020 US presidential election.
- 26 September 2022
- Published by Springer Nature in Information Systems Frontiers
- Vol. 25 (2) , 1-20
Does Fake News in Different Languages Tell the Same Story? An Analysis of Multi-level Thematic and Emotional Characteristics of News about COVID-19
Fake news is being generated in different languages, yet existing studies are dominated by English news. The analysis of fake news content has focused on lexical and stylometric features, giving little attention to semantic features. A few studies involving semantic features have either used them as the inputs to classifiers with no interpretations, or treated them in isolation. This research aims to investigate both thematic and emotional characteristics of fake news at different levels and compare them between different languages for the first time. It extends a state-of-the-art topic modeling technique to extract news topics and introduces a divergence measure to assess the importance of thematic characteristics for identifying fake news. We further examine associations of the thematic and emotional characteristics of fake news. The empirical findings have implications for developing both general and language-specific countermeasures for fake news.
- 26 September 2022
- Published by Springer Nature in Information Systems Frontiers
- Vol. 25 (2) , 1-20
Does Fake News in Different Languages Tell the Same Story? An Analysis of Multi-level Thematic and Emotional Characteristics of News about COVID-19
Fake news is being generated in different languages, yet existing studies are dominated by English news. The analysis of fake news content has focused on lexical and stylometric features, giving little attention to semantic features. A few studies involving semantic features have either used them as the inputs to classifiers with no interpretations, or treated them in isolation. This research aims to investigate both thematic and emotional characteristics of fake news at different levels and compare them between different languages for the first time. It extends a state-of-the-art topic modeling technique to extract news topics and introduces a divergence measure to assess the importance of thematic characteristics for identifying fake news. We further examine associations of the thematic and emotional characteristics of fake news. The empirical findings have implications for developing both general and language-specific countermeasures for fake news.
- 18 March 2022
- Published by Wiley in Journal of the Association for Information Science and Technology
- Vol. 74 (6) , 685-700
Making sense of the black‐boxes: Toward interpretable text classification using deep learning models
Text classification is a common task in data science. Despite the superior performances of deep learning based models in various text classification tasks, their black‐box nature poses significant challenges for wide adoption. The knowledge‐to‐action framework emphasizes several principles concerning the application and use of knowledge, such as ease‐of‐use, customization, and feedback. With the guidance of the above principles and the properties of interpretable machine learning, we identify the design requirements for and propose an interpretable deep learning (IDeL) based framework for text classification models. IDeL comprises three main components: feature penetration, instance aggregation, and feature perturbation. We evaluate our implementation of the framework with two distinct case studies: fake news detection and social question categorization. The experiment results provide evidence for the efficacy of IDeL components in enhancing the interpretability of text classification models. Moreover, the findings are generalizable across binary and multi‐label, multi‐class classification problems. The proposed IDeL framework introduce a unique iField perspective for building trusted models in data science by improving the transparency and access to advanced black‐box models.
- 18 March 2022
- Published by Wiley in Journal of the Association for Information Science and Technology
- Vol. 74 (6) , 685-700
Making sense of the black‐boxes: Toward interpretable text classification using deep learning models
Text classification is a common task in data science. Despite the superior performances of deep learning based models in various text classification tasks, their black‐box nature poses significant challenges for wide adoption. The knowledge‐to‐action framework emphasizes several principles concerning the application and use of knowledge, such as ease‐of‐use, customization, and feedback. With the guidance of the above principles and the properties of interpretable machine learning, we identify the design requirements for and propose an interpretable deep learning (IDeL) based framework for text classification models. IDeL comprises three main components: feature penetration, instance aggregation, and feature perturbation. We evaluate our implementation of the framework with two distinct case studies: fake news detection and social question categorization. The experiment results provide evidence for the efficacy of IDeL components in enhancing the interpretability of text classification models. Moreover, the findings are generalizable across binary and multi‐label, multi‐class classification problems. The proposed IDeL framework introduce a unique iField perspective for building trusted models in data science by improving the transparency and access to advanced black‐box models.