Benchmarking the Robustness of Phishing Email Detection Systems

2023 | Americas Conference on Information Systems | Citations: 0

Authors: Ampel, Benjamin; Gao, Yang; Hu, James; Samtani, Sagar; Chen, Hsinchun

Abstract: Social engineering attacks are currently the most cited cybersecurity threat to ... Expand

Abstract: Social engineering attacks are currently the most cited cybersecurity threat to organizations. Phishing emails are the most salient form of social engineering attacks. Organizations are increasingly implementing AI-enabled systems to detect phishing emails. However, AI-enabled systems are often susceptible to textual perturbations, where an adversary makes a small change to cause a misclassification. In this study, we sought to identify the performance of prevailing phishing email detection systems (PEDS) against character, word, sentence, and multi-level adversarial text perturbations. Through a principled benchmarking framework, we quantitatively demonstrated the lack of robustness prevailing PEDS have to specific types of text-based adversarial perturbations (e.g., character, word, sentence, multi-level). The results of this study provide new insights into the robustness of AI-based PEDS and highlight the need for organizations to adopt a multi-layered approach to phishing protection. Additionally, organizations can implement our benchmark framework to test their PEDS against adversarial perturbations. Collapse

Topics: artificial intelligence electronic mail email phishing phishing IT security

Methods: word2vec random forest classification naïve bayes literature study pre-trained language model

Counterfactual Explanations for Incorrect Predictions Made by AI Models

2023 | International Conference on Information Systems | Citations: 0

Authors: Asrzad, Amir; Li, Xiaobai

Abstract: Advanced AI models are powerful in making accurate predictions for complex prob ... Expand

Abstract: Advanced AI models are powerful in making accurate predictions for complex problems. However, these models often operate as black boxes. This lack of interpretability poses significant challenges, especially in high-stakes applications such as finance, healthcare, and criminal justice. Explainable AI seeks to address the challenges by developing methods that can provide meaningful explanations for humans to understand. When black box models are used for prediction, they inevitably produce errors. It is important to appropriately explain incorrect predictions. This problem, however, has not been addressed in the literature. In this study, we propose a novel method to provide explanations for misclassified cases made by black box models. The proposed method takes a counterfactual explanation approach. It builds a decision tree to find the best counterfactual examples for explanations. Incorrect predictions are rectified using a trust score measure. We validate the proposed method in an evaluation study using real-world data. Collapse

Topics: artificial intelligence explainable artificial intelligence local interpretable model agnostic explanation decision making enterprise information system

Methods: machine learning decision tree classification literature study random forest classification experiment

Rumor Classification through a Multimodal Fusion Framework and Ensemble Learning

2023 | Information Systems Frontiers | Citations: 0

Authors: Azri, Abderrazek; Favre, Cécile; Harbi, Nouria; Darmont, Jérôme; Noûs, Camille

Abstract: The proliferation of rumors on social media has become a major concern due to it ... Expand

Abstract: The proliferation of rumors on social media has become a major concern due to its ability to create a devastating impact. Manually assessing the veracity of social media messages is a very time-consuming task that can be much helped by machine learning. Most message veracity verification methods only exploit textual contents and metadata. Very few take both textual and visual contents, and more particularly images, into account. Moreover, prior works have used many classical machine learning models to detect rumors. However, although recent studies have proven the effectiveness of ensemble machine learning approaches, such models have seldom been applied. Thus, in this paper, we propose a set of advanced image features that are inspired from the field of image quality assessment, and introduce the Multimodal fusiON framework to assess message veracIty in social neTwORks (MONITOR), which exploits all message features by exploring various machine learning models. Moreover, we demonstrate the effectiveness of ensemble learning algorithms for rumor detection by using five metalearning models. Eventually, we conduct extensive experiments on two real-world datasets. Results show that MONITOR outperforms state-of-the-art machine learning baselines and that ensemble models significantly increase MONITOR’s performance. Collapse

Topics: Twitter social network Python microblog logistics management

Methods: computational algorithm machine learning experiment meta learning ensemble learning

Optimizing Substance Use Treatment Selection Using Reinforcement Learning

2023 | ACM Transactions on Management Information Systems | Citations: 0

Authors: Baucum, Matt; Khojandi, Anahita; Myers, Carole; Kessler, Larry

Abstract: Substance use disorder (SUD) exacts a substantial economic and social cost in th ... Expand

Abstract: Substance use disorder (SUD) exacts a substantial economic and social cost in the United States, and it is crucial for SUD treatment providers to match patients with feasible, effective, and affordable treatment plans. The availability of large SUD patient datasets allows for machine learning techniques to predict patient-level SUD outcomes, yet there has been almost no research on whether machine learning can be used to optimize or personalize which treatment plans SUD patients receive. We use contextual bandits (a reinforcement learning technique) to optimally map patients to SUD treatment plans, based on dozens of patient-level and geographic covariates. We also use near-optimal policies to incorporate treatments’ time-intensiveness and cost into our recommendations, to aid treatment providers and policymakers in allocating treatment resources. Our personalized treatment recommendation policies are estimated to yield higher remission rates than observed in our original dataset, and they suggest clinical insights to inform future research on data-driven SUD treatment matching. Collapse

Topics: electronic health record individual characteristic healthcare data missing data

Methods: hazard function machine learning random forest classification descriptive statistic propensity score method

Expl(AI)ned: The Impact of Explainable Artificial Intelligence on Users’ Information Processing

2023 | Information Systems Research | Citations: 0

Authors: Bauer, Kevin; von Zahn, Moritz; Hinz, Oliver

Abstract: Because of a growing number of initiatives and regulations, predictions of moder ... Expand

Abstract: Because of a growing number of initiatives and regulations, predictions of modern artificial intelligence (AI) systems increasingly come with explanations about why they behave the way they do. In this paper, we explore the impact of feature-based explanations on users’ information processing. We designed two complementary empirical studies where participants either made incentivized decisions on their own, with the aid of opaque predictions, or with explained predictions. In Study 1, laypeople engaged in the deliberately abstract investment game task. In Study 2, experts from the real estate industry estimated listing prices for real German apartments. Our results indicate that the provision of feature-based explanations paves the way for AI systems to reshape users’ sense making of information and understanding of the world around them. Specifically, explanations change users’ situational weighting of available information and evoke mental model adjustments. Crucially, mental model adjustments are subject to the confirmation bias so that misconceptions can persist and even accumulate, possibly leading to suboptimal or biased decisions. Additionally, mental model adjustments create spillover effects that alter user behavior in related yet disparate domains. Overall, this paper provides important insights into potential downstream consequences of the broad employment of modern explainable AI methods. In particular, side effects of mental model adjustments present a potential risk of manipulating user behavior, promoting discriminatory inclinations, and increasing noise in decision making. Our findings may inform the refinement of current efforts of companies building AI systems and regulators that aim to mitigate problems associated with the black-box nature of many modern AI systems. History: Alessandro Acquisti, senior editor; Jason Chan, associate editor. Funding: This work was supported by the Deutsche Forschungsgemeinschaft (DFG) (Projek 449023539), Volkswagen Foundation (ML2MT), and LeibnizInstitute for Financial Research SAFE. Supplemental Material: The online appendix is available at https://doi.org/10.1287/isre.2023.1199. Collapse

Topics: explainable artificial intelligence artificial intelligence shapley additive explanation local interpretable model agnostic explanation decision making

Methods: F-test regression analysis method experiment field study robustness check

Human-machine collaboration for feature selection and integration to improve congestive Heart failure risk prediction

2023 | Decision Support Systems | Citations: 0

Authors: Ben-Assuli, Ofir; Heart, Tsipi; Klempfner, Robert; Padman, Rema

Abstract: The issue of harnessing machine learning (ML) algorithms for the prediction of a ... Expand

Abstract: The issue of harnessing machine learning (ML) algorithms for the prediction of adverse medical events is important considering the prevalence of vast repositories of patient-level medical data, and the quest to reduce healthcare costs while elevating the quality of care. Of particular interest is identifying a set of features that can better predict the risk of unplanned readmissions of chronically ill patients within 30 days post discharge, among them patients with Congestive Heart Failure (CHF). While numerous studies have employed ML algorithms to identify sets of significant features, only a handful have compared features extracted by various methods, and even fewer have compared the predictive power of feature sets selected by human experts to those selected by ML. Hence, this research aimed to elicit a comprehensive feature set selected by four ML algorithms and compare its predictive performance to that of a feature set selected by experts. We then evaluated a merged list of the two feature sets, constructing a human-machine collaborative set to predict the likelihood of 30-day unplanned readmission of CHF patients using data on 10,763 CHF patients hospitalized during the years 2010–2017. Our models achieved Area under the ROC Curve (AUC) above 0.8 and an Accuracy of ~0.89, comparable to the best models in the extant literature. Moreover, the Human-Machine Collaborative model significantly outperformed (best AUC 0.836) other human-only selected models as well as the machine-selected set. This study contributes to a better understanding of the power of using ML for patient risk stratification with special attention to the benefits of human-machine collaboration, even when these two entities work separately or in parallel, for improved clinical decision making. Collapse

Topics: healthcare data decision making big data missing data data warehouse

Methods: machine learning computational algorithm XGBoost descriptive statistic random forest classification

Save Our Souls: Study of Twitter Use during India’s COVID-19 Pandemic

2023 | International Conference on Information Systems | Citations: 0

Authors: Bhattacharyya, Samadrita; Mojumder, Probal

Abstract: Twitter is a commonly used social platform for communication during disasters. ... Expand

Abstract: Twitter is a commonly used social platform for communication during disasters. Tweets by citizens during disasters to share information, seek, and offer help create a body of spontaneous, decentralized, emergent social media communication. Users' exploit Twitter's reach-enabling technological functionalities (hashtags (#), mentions (@), and 'reply-to') to draw attention to the messages. Set in context of the second wave of COVID-19 in India, that saw a surge in citizen-driven tweets seeking healthcare resources from fellow citizens and officials (i.e., SOS tweets), our paper empirically analyses the effects of Twitter's reach-enabling functionalities on online responses (i.e., retweets and replies) to these SOS tweets. We investigate the effects of inclusion of hashtags, mentions, and 'reply to' SOS tweets. We also examine the moderating effect of how the effects of the reach-enabling functionalities change as the social platform gets crowded with SOS tweets. The study offers various academic and practical implications. Collapse

Topics: Twitter pandemic social media Python societal evaluation criteria

Methods: propensity score method machine learning propensity score matching econometric modeling robustness check

AI-Enhanced Hybrid Decision Management

2023 | Business & Information Systems Engineering | Citations: 0

Authors: Bork, Dominik; Ali, Syed Juned; Dinev, Georgi Milenov

Abstract: The Decision Model and Notation (DMN) modeling language allows the precise speci ... Expand

Abstract: The Decision Model and Notation (DMN) modeling language allows the precise specification of business decisions and business rules. DMN is readily understandable by business users involved in decision management. However, as the models get complex, the cognitive abilities of humans threaten manual maintainability and comprehensibility. Proper design of the decision logic thus requires comprehensive automated analysis of e.g., all possible cases the decision shall cover; correlations between inputs and outputs; and the importance of inputs for deriving the output. In the paper, the authors explore the mutual benefits of combining human-driven DMN decision modeling with the computational power of Artificial Intelligence for DMN model analysis and improved comprehension. The authors propose a model-driven approach that uses DMN models to generate Machine Learning (ML) training data and show, how the trained ML models can inform human decision modelers by means of superimposing the feature importance within the original DMN models. An evaluation with multiple real DMN models from an insurance company evaluates the feasibility and the utility of the approach. Collapse

Topics: artificial intelligence explainable artificial intelligence perceived usefulness price management extensible markup language

Methods: computational algorithm decision tree classification experiment conceptual modelling simulation

Augmenting Social Bot Detection with Crowd-Generated Labels

2023 | Information Systems Research | Citations: 0

Authors: Benjamin, Victor; Raghu, T. S.

Abstract: Social media platforms are facing increasing numbers of cyber-adversaries seekin ... Expand

Abstract: Social media platforms are facing increasing numbers of cyber-adversaries seeking to manipulate online discourse by using social bots (i.e., social media software robots) to help automate and scale their attacks. Likewise, some social media users can identify social bot activity at varying degrees of confidence. In this research, human reactions to social bot messages are used to augment existing social bot detection capabilities. Speech act theory is used to inspire a framework for assessing the credibility of instances where users identify potential bot activity, as not all user responses are of equal credibility for assisting with the bot detection task. The framework is then operationalized through deep learning methodologies to develop a computational system for identifying social bots. Real-world performance and practicality of the developed framework is demonstrated on a live, crowd-sourced data set collected from a real-world social media platform. Results show that consideration of crowd reactions to suspected bots can significantly improve bot detection performance. Furthermore, consideration of speech acts to evaluate crowd reactions can even further augment the system’s performance, although speech acts themselves are not necessary to observe performance boost through crowd intelligence. This study serves as a grounding point for future work that can explore an augmented model for detecting other forms of algorithmically generated content within social media platforms.History: Olivia Sheng, Senior Editor; Jesse Bockstedt, Associate Editor.Supplemental Material: The online appendix is available at https://doi.org/10.1287/isre.2022.1136. Collapse

Topics: Reddit social media crowdsourcing IT productivity social media analytics

Methods: experiment BERT machine learning support vector machine random forest classification

Theories: speech act theory

Deploying Artificial Intelligence to Combat Covid-19 Misinformation on Social Media: Technological and Ethical Considerations

2023 | HICSS | Citations: 0

Authors: Cartwright, B.; Frank, R.; Weir, G.; Padda, K.; Strange, S.

Abstract: This paper reports on AI research into online misinformation pertaining to the ... Expand

Abstract: This paper reports on AI research into online misinformation pertaining to the COVID-19 pandemic within the Canadian context. This is part of our longerterm goal, i.e., development of a machine-learning tool to assist social media platforms, online service providers and government agencies in identifying and responding to misinformation on social media. We report on predictive accuracies accomplished by applying a combination of technologies, including a custom-designed web-crawler, The Dark Crawler, the Posit toolkit, and four different machine-learning models based on Naïve Bayes, Support Vector Machines, LibLinear and LibShortText. Overall, we found that Posit and LibShortText models showed higher levels of correlation to the pre-determined (manual and machine-driven) data classifications than the other machine-learning algorithms tested. We further argue that the harms associated with COVID-19 misinformatione.g., the social and economic damage, and the deaths and severe illnessesoutweigh the right to personal privacy and freedom of speech considerations. Collapse

Topics: misinformation Twitter Reddit social media pandemic

Methods: machine learning computational algorithm support vector machine decision tree classification random forest classification