Knowledge that Transforms

The prevalence and rapid growth of cybercrime are largely attributed to hacker communities on the dark web, where cybercriminals extensively exchange hacking resources, share hacking knowledge, and organize cyberattacks. Such streams of hacker-generated content constitute an invaluable data source for developing threat intelligence that can inform organizations of cybersecurity risks and facilitate proactive cyber defense. Drawing upon the design science paradigm, we propose a novel nonparametric emerging topic detection (NPETD) framework for detecting emerging topics in streams of hacker-generated content. Our framework extends the state-of-the-art nonparametric topic model to inductively model topics without having to specify the number of topics a priori. Moreover, our framework features an efficient algorithm to jointly infer topics and detect topic emergence. We conducted experiments to rigorously evaluate the effectiveness and efficiency of our framework in comparison with the state-of-the-art baseline methods. Our framework outperformed the baseline methods in detecting the listings of emerging threats in darknet marketplaces on recall, F-measure, topic coherence, and processor time. The practical utility of our framework is further demonstrated in a major hacker forum, where we identified several notable emerging topics with important implications for victim companies and law enforcement. The proposed framework contributes to cybersecurity, topic detection and tracking, and design science.

Online social cues that utilize user-generated data, such as user reviews and product ratings, have become one of the key factors influencing online user behavior and decisions. Online users who shared their reviews and ratings about a product (or a seller) become an abstract reference group to a focal user interested in the same product. This study focuses on sponsored search results (SSRs), a type of unsolicited information that matches users’ search queries and receives high evaluations from prior consumers. We investigate the effects of positive social cues on alleviating users’ avoidance responses toward an encountered SSR when searching for a product in a C2C e-commerce context. We synthesize the avoidance literature and identify three forms of SSR avoidance, namely, cognitive, behavioral, and affective avoidance. We apply users’ implicit concerns on SSRs to explain users’ avoidance of an encountered SSR. In addition, we extend social influence theory to online settings where abstract reference groups are posited to trigger social influence. We examine how and under what conditions the three forms of SSR avoidance can be reduced by various positive online social cues (i.e., product- and seller-related). We conduct three laboratory experiments. Results attest to users’ implicit concerns on SSRs and their avoidance of SSRs and reveal different effects of various social cues on reducing the three forms of SSR avoidance. This study uncovers the theoretical mechanisms of social influence on reducing SSR avoidance in online settings. It also offers practical implications for online search service providers to help online users’ decision making in their search process.

Weak passwords are one of the most pervasive threats in cybersecurity. Facing this threat, users require guidance on how to protect themselves. A method frequently used by IS practitioners and researchers to provide this guidance is fear appeals, persuasive messages intended to prompt behavioral changes in response to a threat. However, previous research has not considered a key element of fear appeal effectiveness: task primacy. When fear appeals are a part of the primary or focal task, users’ cognitive engagement will be high by default. However, when fear appeals are delivered as secondary tasks, such as interruptive security messages, users’ engagement is likely to be low because the primary task takes priority in attentional and cognitive resources. In such cases, a remedy is needed to elicit engagement with the fear appeal. In this research note, we theorize that cognitive engagement acts as a contextual moderator that is critical to the effectiveness of fear appeals under the boundary condition of task primacy. Further, we theorize that interactivity, a mechanism that adapts message content through tailored real-time feedback in response to a user’s actions, is a key remedy to enhance engagement with fear appeals. However, to date fear appeals have largely been tested in noninteractive primary tasks, and no study has provided a theoretical explanation for why interactivity enhances the power of a fear appeal. We empirically examined engagement as a contextual moderator in two ways. First, we conducted a field experiment, which manipulated messages on a password creation form on a real-world website. Second, we performed a qualitative focus group study to triangulate the experimental results and more fully reify our theoretical model. Together, the findings reveal that interactivity acts as a catalyst to engage participants with a fear appeal, which then allows the persuasive message of the fear appeal to be internalized. The concepts of boundary condition of task primacy and engagement suggest ways that fear appeals can be more effectively applied in research and practice.

Findings on how prior high-quality ideas affect the quality of subsequent ideas in online ideation contests have been mixed. Some studies find that high-quality ideas lead to subsequent high-quality ideas, while others find the opposite. Based on computationally intensive exploratory research, utilizing theory on blending of mental spaces, we suggest that the effects of prior ideas on the generation of subsequent ideas depend on the alignment of (1) crowd participants’ subjective quality assessments of prior ideas and (2) subsequent problem-related contributions made by the crowd. When a prior idea is assessed as high-quality, this motivates the crowd to emulate that idea. When this motivation is aligned with subsequent contributions that expand the mental space of the prior idea, a new high-quality idea can be created. In contrast, when a prior idea is assessed as low-quality, it motivates the crowd to redirect away from that idea. When this motivation is aligned with subsequent contributions that shift the mental space of the prior idea, a new high-quality idea can be created. The mixed findings in the literature can then be explained by a failure to consider non-idea information contributions made by the crowd.

In this paper, we examine the effectiveness of various recommendation strategies in the mobile channel and their impact on consumers’ utility and demand levels for individual products. We find significant differences in effectiveness among various recommendation strategies. Interestingly, recommendation strategies that directly embed social proofs for the recommended alternatives outperform other recommendations. In addition, recommendation strategies combining social proofs with higher levels of induced awareness due to the prescribed temporal diversity have an even stronger effect on the mobile channel. We also examine the heterogeneity of the demand effect across items, users, and contextual settings, further verifying empirically the aforementioned information and persuasion mechanisms and generating rich insights. We also facilitate the estimation of causal effects in the presence of endogeneity using machine-learning methods. Specifically, we develop novel econometric instruments that capture product differentiation (isolation) based on deep-learning models of user-generated reviews. Our empirical findings extend the current knowledge regarding the heterogeneous impact of recommender systems, reconcile contradictory prior results in the related literature, and have significant business implications.

In this paper, we examine the role of reciprocity enabled by digital social platforms for offline healthy behavior. Although reciprocity is a fundamental aspect of human psychology, its application in promoting healthy behavior has been limited. We conduct a randomized field experiment with over 1,700 pairs of users on a mobile social network platform. Individuals in the reciprocity treatment group receive a gift from their friends and are asked to return this favor by participating in a running challenge. Their performance is compared to the self-interest incentives widely used in practice. Building on social exchange theory, we argue that reciprocity-based incentives will outperform self-interest incentives with modest reward for motivating behavior change. We find that, on average, reciprocity is stronger than self-interest in inducing exercise behavior by a substantial amount. Furthermore, our results reveal that the magnitude of the reciprocity effect is contingent on the social closeness between senders and receivers. Interestingly, social closeness has an inverted U-shaped influence on the reciprocity effect. The effect is strongest when closeness is moderate, and wanes when closeness is either too strong or too weak. Compared to commonly used self-interest based financial incentives, our findings offer a potentially more powerful avenue for mechanism design in promoting healthy behavior.

Information technology (IT) professionals are a strategic human resource for enabling competitive advantage through the application of data and technologies. Yet, it remains a challenge for organizations to retain top IT talent as the business context and the nature of work change. Retention strategies that have worked with other business professionals have faced limited success with IT talent, leading some scholars to ask whether the latter are unique in terms of culture, personality, and tools. This study seeks to address this question by (1) undertaking a meta-analysis of studies investigating the turnover intentions of IT professionals, and by (2) conducting a second-order meta-analysis to compare findings between IT and non-IT professionals. The meta-analytic findings identify new antecedents more recently examined in information systems (IS) research while confirming enduring relationships between antecedents and turnover intention. The second-order meta-analysis provides intriguing findings regarding the potential uniqueness of IT professionals. We provide an integrative discussion on the state of turnover intention research within the IS discipline, start a dialog on interdisciplinary comparisons, and offer a forward-looking agenda for future research. We conclude by calling on scholars to revitalize turnover intention research within the IS field by moving toward fresh theories, fresh constructs, and fresh approaches.

Advancing the quality of healthcare for senior citizens with chronic conditions is of great social relevance. To better manage chronic conditions, objective, convenient, and inexpensive wearable sensor-based information systems (IS) have been increasingly used by researchers and practitioners. However, existing models often focus on a single aspect of chronic conditions and are often “black boxes” with limited interpretability. In this research, we adopt the computational design science paradigm and propose a novel adversarial attention-based deep multisource multitask learning (AADMML) framework. Drawing upon deep learning, multitask learning, multisource learning, attention mechanism, and adversarial learning, AADMML addresses limitations with existing wearable sensor-based chronic condition severity assessment methods. Choosing Parkinson’s disease (PD) as our test case because of its prevalence and societal significance, we conduct benchmark experiments to evaluate AADMML against state-of-the-art models on a large-scale dataset containing thousands of instances. We present three case studies to demonstrate the practical utility and economic benefits of AADMML and by applying it to detect early-stage PD. We discuss how our work is related to the IS knowledge base and its practical implications. This work can contribute to improved life quality for senior citizens and advance IS research in mobile health analytics.

Trace data—users’ digital records when interacting with technology—can reveal their cognitive dynamics when making decisions on websites in real time. Here, we present a trace-data method, analyzing movements captured via a computer mouse, to assess potential fraud when filling out an online form. In contrast to existing fraud-detection methods, which analyze information after submission, mouse-movement traces can capture the cognitive deliberations as possible indicators of fraud as it is happening. We report two controlled studies using different tasks, where participants could freely commit fraud to benefit themselves financially. As they performed the tasks, we captured mouse-cursor movement data and found that participants who entered fraudulent responses moved their mouse significantly more slowly and with greater deviation. We show that the extent of fraud matters such that more extensive fraud increases movement deviation and decreases movement speed. These results demonstrate the efficacy of analyzing mouse-movement traces to detect fraud during online transactions in real time, enabling organizations to confront fraud proactively as it is happening at internet scale. Our method of analyzing actual user behaviors in real time can complement other behavioral methods in the context of fraud and a variety of other contexts and settings.

This paper reviews the literature on how organizations learn from information system (IS) incidents. We identify three modes of learning depending on the practices that constitute the learning process, the specific actors who play roles in learning, the temporal orientation of the learning practices, and the specific contextual focus of the learning. The literature focuses primarily on learning from past experience to draw lessons for future incidents (reflective learning mode). Yet, a growing stream of literature stresses the importance of learning through engagement with present incidents (embedded learning mode), and a few studies suggest that organizations can learn prospectively to prepare for future incidents (prospective learning mode). We argue that although these three learning modes are effective, they do not adequately explain how organizations learn from IS incidents when used in isolation. Since IS incidents unfold increasingly as sets of interacting events across information systems and organizational settings, organizational learning needs to be theorized as an iterative process among these learning modes. We synthesize these three learning modes into an integrative framework and theorize about their supportive and inhibiting relations. We suggest some opportunities for future research, which would advance our understanding of how organizations learn from IS incidents.