Providing insights into health data science education through artificial intelligence.

BACKGROUND: Health Data Science (HDS) is a novel interdisciplinary field that integrates biological, clinical, and computational sciences with the aim of analysing clinical and biological data through the utilisation of computational methods. Training healthcare specialists who are knowledgeable in both health and data sciences is highly required, important, and challenging. Therefore, it is essential to analyse students' learning experiences through artificial intelligence techniques in order to provide both teachers and learners with insights about effective learning strategies and to improve existing HDS course designs. METHODS: We applied artificial intelligence methods to uncover learning tactics and strategies employed by students in an HDS massive open online course with over 3,000 students enrolled. We also used statistical tests to explore students' engagement with different resources (such as reading materials and lecture videos) and their level of engagement with various HDS topics. RESULTS: We found that students in HDS employed four learning tactics, such as actively connecting new information to their prior knowledge, taking assessments and practising programming to evaluate their understanding, collaborating with their classmates, and repeating information to memorise. Based on the employed tactics, we also found three types of learning strategies, including low engagement (Surface learners), moderate engagement (Strategic learners), and high engagement (Deep learners), which are in line with well-known educational theories. The results indicate that successful students allocate more time to practical topics, such as projects and discussions, make connections among concepts, and employ peer learning. CONCLUSIONS: We applied artificial intelligence techniques to provide new insights into HDS education. Based on the findings, we provide pedagogical suggestions not only for course designers but also for teachers and learners that have the potential to improve the learning experience of HDS students.

Predicting suicide risk in real-time crisis hotline chats integrating machine learning with psychological factors: Exploring the black box.

BACKGROUND: This study addresses the suicide risk predicting challenge by exploring the predictive ability of machine learning (ML) models integrated with theory-driven psychological risk factors in real-time crisis hotline chats. More importantly, we aimed to understand the specific theory-driven factors contributing to the ML prediction of suicide risk. METHOD: The dataset consisted of 17,654 crisis hotline chat sessions classified dichotomously as suicidal or not. We created a suicide risk factors-based lexicon (SRF), which encompasses language representations of key risk factors derived from the main suicide theories. The ML model (Suicide Risk-Bert; SR-BERT) was trained using natural language processing techniques incorporating the SRF lexicon. RESULTS: The results showed that SR-BERT outperformed the other models. Logistic regression analysis identified several theory-driven risk factors significantly associated with suicide risk, the prominent ones were hopelessness, history of suicide, self-harm, and thwarted belongingness. LIMITATIONS: The lexicon is limited in its ability to fully encompass all theoretical concepts related to suicide risk, nor to all the language expressions of each concept. The classification of chats was determined by trained but non-professionals in metal health. CONCLUSION: This study highlights the potential of how ML models combined with theory-driven knowledge can improve suicide risk prediction. Our study underscores the importance of hopelessness and thwarted belongingness in suicide risk and thus their role in suicide prevention and intervention.

Supervisor-Worker Problems with an Application in Education.

In many e-learning settings, allowing students to choose which skills to practice encourages their motivation and contributes to learning. However, when given choice, students may prefer to practice skills that they already master, rather than practice skills they need to master. On the other hand, requiring students only to practice their required skills may reduce their motivation and lead to dropout. In this paper, we model this tradeoff as a multi-agent planning task, which we call SWOPP (Supervisor- Worker Problem with Partially Overlapping goals), involving two agents-a supervisor (teacher) and a worker (student)-each with different, yet non-conflicting, goals. The supervisor and worker share joint goals (mastering skills). The worker plans to achieve his/her own goals (completing an e-learning session) at a minimal cost (effort required to solve problems). The supervisor guides the worker towards achieving the joint goals by controlling the problems in the choice set for the worker. We provide a formal model for the SWOPP task and two sound and complete algorithms for the supervisor to guide the worker's plan to achieve their joint goals. We deploy SWOPP for the first time in a real-world study to personalize math questions for K5 students using an e-learning software in schools. We show that SWOPP was able to guide students' interactions with the software to practice necessary skills without deterring their motivation.

Comparing Plan Recognition Algorithms Through Standard Plan Libraries.

Plan recognition deals with reasoning about the goals and execution process of an actor, given observations of its actions. It is one of the fundamental problems of AI, applicable to many domains, from user interfaces to cyber-security. Despite the prevalence of these approaches, they lack a standard representation, and have not been compared using a common testbed. This paper provides a first step towards bridging this gap by providing a standard plan library representation that can be used by hierarchical, discrete-space plan recognition and evaluation criteria to consider when comparing plan recognition algorithms. This representation is comprehensive enough to describe a variety of known plan recognition problems and can be easily used by existing algorithms in this class. We use this common representation to thoroughly compare two known approaches, represented by two algorithms, SBR and Probabilistic Hostile Agent Task Tracker (PHATT). We provide meaningful insights about the differences and abilities of these algorithms, and evaluate these insights both theoretically and empirically. We show a tradeoff between expressiveness and efficiency: SBR is usually superior to PHATT in terms of computation time and space, but at the expense of functionality and representational compactness. We also show how different properties of the plan library affect the complexity of the recognition process, regardless of the concrete algorithm used. Lastly, we show how these insights can be used to form a new algorithm that outperforms existing approaches both in terms of expressiveness and efficiency.

EEG-Based Prediction of Cognitive Load in Intelligence Tests.

Measuring and assessing the cognitive load associated with different tasks is crucial for many applications, from the design of instructional materials to monitoring the mental well-being of aircraft pilots. The goal of this paper is to utilize EEG to infer the cognitive workload of subjects during intelligence tests. We chose the well established advanced progressive matrices test, an ideal framework because it presents problems at increasing levels of difficulty and has been rigorously validated in past experiments. We train classic machine learning models using basic EEG measures as well as measures of network connectivity and signal complexity. Our findings demonstrate that cognitive load can be well predicted using these features, even for a low number of channels. We show that by creating an individually tuned neural network for each subject, we can improve prediction compared to a generic model and that such models are robust to decreasing the number of available channels as well.

The more the merrier? Increasing group size may be detrimental to decision-making performance in nominal groups.

Demonstrability-the extent to which group members can recognize a correct solution to a problem-has a significant effect on group performance. However, the interplay between group size, demonstrability and performance is not well understood. This paper addresses these gaps by studying the joint effect of two factors-the difficulty of solving a problem and the difficulty of verifying the correctness of a solution-on the ability of groups of varying sizes to converge to correct solutions. Our empirical investigations use problem instances from different computational complexity classes, NP-Complete (NPC) and PSPACE-complete (PSC), that exhibit similar solution difficulty but differ in verification difficulty. Our study focuses on nominal groups to isolate the effect of problem complexity on performance. We show that NPC problems have higher demonstrability than PSC problems: participants were significantly more likely to recognize correct and incorrect solutions for NPC problems than for PSC problems. We further show that increasing the group size can actually decrease group performance for some problems of low demonstrability. We analytically derive the boundary that distinguishes these problems from others for which group performance monotonically improves with group size. These findings increase our understanding of the mechanisms that underlie group problem-solving processes, and can inform the design of systems and processes that would better facilitate collective decision-making.