Asking appropriate follow-up questions is an important aspect of semi-structured interviews, yet interviewers’ knowledge and skills may limit their depth and affect outcomes. While many studies have demonstrated the usefulness of large language models (LLMs) for qualitative analysis, their potential in data collection remains underexplored. We adopt an AI-driven/Reverse “Wizard-of-Oz” setup to investigate how real-time LLM support for generating follow-up questions shapes semi-structured interviews. Through a study with 17 participants, we examine the value of LLM-generated follow-up questions, the evolving division of roles, relationships, collaborative behaviors, and responsibilities between interviewers and AI. Our findings (1) provide empirical evidence of the strengths and limitations of AI-generated follow-up questions (AGQs); (2) introduce a Human-AI collaboration framework in this interview context; and (3) propose human-centered design guidelines for AI-assisted interviews. We position LLMs as complements, not replacements, human judgment, and highlight pathways for integrating AI into qualitative data collection.
Abstract Traditional two-dimensional (2D) hydraulic models face significant computational challenges (e.g., long computing times) that limit their use in applications that are time-sensitive or require many model evaluations. This study presents a physics-informed deep-operator network (DeepONet) framework for computational hydraulics modeling that learns the solution operator of the 2D shallow water equations (SWEs) to create fast surrogate models. The framework can operate in two modes: a purely data-driven SWE-DeepONet that learns from a numerical solver such as Sedimentation and River Hydraulics–Two-Dimension (SRH-2D), and a physics-informed shallow water equations DeepONet (PI-SWE-DeepONet) that additionally incorporates the continuous SWEs as constraints during training. Based on a real-world case, steady flows in a reach of the Sacramento River in California, it is demonstrated that PI-SWE-DeepONet possesses much enhanced prediction capability than SWE-DeepONet when applied to out-of-distribution scenarios. The physics-informed model is shown to exhibit slower error growth and larger breakdown distances in the normalized inlet-discharge parameter space (the distance between an out-of-distribution case and the closest training sample) in comparison with SWE-DeepONet. The gain of the physics-informed training, however, comes with costs, chief among which is that the simulated results have slightly higher errors for in-distribution cases. It reflects the existence of a tension between the two competing training objectives: fitting the results from the traditional hydraulic model and satisfying the continuous governing equations. In this study, guidelines are developed for selecting the appropriate approach based on a real-world case: PI-SWE-DeepONet is preferred for out-of-distribution predictions, uncertain training data, or when physical consistency is a priority, while SWE-DeepONet is recommended if the modeling objective is to replicate faithfully the traditional hydraulic model results within the training distribution. Once trained, both surrogates evaluate a new scenario in less than one second on a single graphics processing unit (GPU), more than two orders of magnitude faster than a fresh SRH-2D run on the same case, which makes them suitable for applications requiring many model evaluations. Other challenges are also discussed, such as the loss weighting approach.
This paper introduces a framework based on Large Deviation Theory (LDT) to accurately and efficiently compute the rare probabilities of voltage collapse. We formulate the problem as finding the most probable failure point (the instanton) on the stability boundary and derive both first-order and second-order approximations for the collapse probability. The second-order method incorporates the local curvature of the stability boundary, yielding higher accuracy. This LDT framework generalizes methods based on Mahalanobis distance and is extensible to non-Gaussian uncertainties. We validate our approach on test systems, demonstrating that the LDT estimates converge to Monte Carlo results in the rare-event regime where direct sampling becomes computationally prohibitive.
Crisis discourse is a significant site for reporting crises and for discursively managing the public’s emotions. Existing studies of this discourse have largely examined how crises are discursively constructed and responded to, with reassurance treated primarily as a function of responses; few have focused on the reassuring potential of the situation itself. This study therefore investigates the discursive strategies through which media reports construe the COVID-19 situation as improving – a form of reassurance grounded in the situation. Using a mixed-methods approach, the data are analyzed with reference to the appraisal framework and the re-/coupling concept in systemic functional linguistics. Situation improvement is achieved discursively by mitigating and inverting negative evaluations of the crisis: negative values are downgraded or recoupled to improvement-oriented targets, which constructs a sense of persistent threat while subtly inviting a more positive reading of the situation. These findings could deepen understanding of the discursive management of crisis and the role of linguistic analysis in illuminating crisis communication and public reassurance in media discourse.
Background Symptoms of many gastrointestinal infections are similar, making it difficult to differentiate the causative agents. Automated PCR-based diagnostic systems can rapidly detect and identify pathogens responsible for these infections. Methods BD Enteric Bacterial Panel plus (EBP plus Assay) for BD COR System was assessed versus a single comparator (BioFire™ FilmArray™ GI Panel) or a composite reference (≥2 of 3 assays: BioFire FilmArray GI Panel, BioCode™ GPP, and BD MAX™ EBP/xEBP) for detecting enteric pathogens using patient stool specimens. Positive and negative percent agreement (PPA and NPA) between EBP plus Assay and comparator method(s) were calculated with 95% two-sided confidence intervals (CIs). Agreement ≥90% with CI lower bounds ≥80% was considered acceptable performance. Contrived specimens were included for low prevalence targets. Results There were 1737 specimens analyzed. NPA met performance thresholds (≥98.4%) for all targets. PPA met performance thresholds for Salmonella spp., Campylobacter spp., and Shiga toxin. Due to low prevalence in the study population, the PPA lower bound of the 95% CI was below the threshold for Shigella spp./enteroinvasive Escherichia coli, Vibrio spp., and Yersinia enterocolitica, while PPA point estimates were not met for Plesiomonas shigelloides and enterotoxigenic E. coli (ETEC). PPA and NPA thresholds were met in contrived specimen testing for Shiga toxin, Vibrio spp., Plesiomonas shigelloides, Yersinia enterocolitica, and ETEC. Conclusions The EBP plus Assay implemented on the BD COR System demonstrated good concordance with the comparator method(s) for detecting enteric bacterial pathogens in stool and is suited for high-throughput laboratories seeking to enhance task automation without impacting accuracy.
