Burcu Arslan

Background. Multiplication is a key skill in primary education, but traditional teaching methods vary in effectiveness and may not suit individual needs. Adaptive learning systems personalise practice and promote the use of effective study strategies, with algorithms that adapt based on accuracy (accuracy-adaptive), or accuracy and speed (RT-adaptive). Incorporating response times should enable more precise adaptivity, leading to better learning. A prior study (n = 540) with both kinds of algorithm showed promising results, but did not compare them directly. Emphasising speed may also affect learners differently, especially those with attentional difficulties. Aims. This study evaluates whether practice with an RT-adaptive spaced repetition algorithm improves multiplication fluency, compared to a simpler accuracy-adaptive algorithm. Sample. Participants were 44 Dutch primary school pupils (grade 4; US grade 2; aged 7–8). The sample included participants with average and raised levels of hyperactivity/inattention.Methods. A within-subject experiment compared practice with accuracy- and RT-adaptive algorithms. We tested multiplication fluency before and after practice (immediately after and after a 1-week delay) and recorded participants’ subjective experiences.Results. Performance was better during RT-adaptive practice, but this difference did not translate to better immediate or delayed post-test outcomes.Hyperactivity/inattention levels did not affect performance but did influence experience: participants with raised levels reported more difficulty concentrating in the RT-adaptive condition.Conclusions. We found limited added value of RT-adaptive practice, but saw that individual differences influenced experiences of AI-supported learning. These findings contribute to a deeper understanding of an effective application of AI in primary education that meets the needs of neurodiverse learners.

In this chapter, we provide an overview of adapting to learners’ prior and developing knowledge in personalized learning (PL) environments from a cognitive science perspective. We first present a theoretical overview by leveraging unified theories of cognition that provide a computational account of how knowledge is represented, how prior knowledge interacts with developing knowledge, and how this interaction affects learning, forgetting, and transfer. Then, we briefly overview the connections between theory and PL applications. Finally, we discuss the benefits and challenges of adapting to prior and developing knowledge and the keys to effective design for PL applications that adapt to prior and developing knowledge.

In line with the positive effects of personalized learning, personalized assessments are expected to maximize learner motivation and engagement, allowing learners to show what they truly know and can do. Considering the advances in Generative Artificial Intelligence (GenAI), in this perspective article, we elaborate on the opportunities of integrating GenAI into personalized educational assessments to maximize learner engagement, performance, and access. We also draw attention to the challenges of integrating GenAI into personalized educational assessments regarding its potential risks to the assessment’s core values of validity, reliability, and fairness. Finally, we discuss possible solutions and future directions.

In educational settings, students rely on metacognitive processes to determine whether or not to exert effort. We investigated ways to minimize cognitively disengaged responses (i.e., not-fully-effortful responses) during a low-stakes mathematics assessment. Initially, we established theory-driven time thresholds for each item to detect such responses. We then administered the test to 800 eighth-graders across three conditions: (a) control (n = 271); (b) instruction (n = 267); and (c) nudge (n = 262). In the instruction condition, students were told to exert their best effort before starting the assessment. In the nudge condition, students were prompted to give their best effort following each first-attempt response that was both incorrect and not-fully-effortful. Therefore, students had multiple opportunities to adjust their level of effort. Nudges, but not effort instruction, significantly reduced students’ not-fully-effortful responses. Neither the nudges nor the effort instruction significantly impacted performance. In a post-test survey, most students reported that they received nudges whenever they did not know the answer (55%). Overall, these findings suggest that while nudges reduce cognitively disengaged responses, most students appear to strategically modulate their level of effort based on self-monitoring their knowledge and response effort.

Pauses represented in process data captured from digital learning and assessment tasks are defined as the time elapsed between two subsequent events. Pauses are used as a marker of unobservable cognitive processes, such as encoding, problem-solving, and planning, that underlie test takers’ subsequent observable actions in educational assessment. To make valid inferences about the underlying cognitive processes represented by pauses, we argue that applying a task-specific cognitive modeling approach is required. We discuss and demonstrate how to apply a task-specific, theoretical cognitive modeling approach for interpreting pauses. We believe that this approach will have value for educational researchers seeking to make valid, task-general inferences about test-taker cognition from pauses represented in process data.

Students exhibit many behaviors when responding to items on a computer-based test, but only some of these behaviors are relevant to estimating their proficiencies. In this study, we analyzed data from computer-based math achievement tests administered to elementary school students in grades 3 (ages 8–9) and 4 (ages 9–10). We investigated students’ response process data, including the total amount of time they spent on an item, the amount of time they took to first respond to an item, the number of times they “visited” an item and the number of times they changed their responses to items, in order to explore whether these behaviors were related to overall proficiency and whether they differed across item formats and grades. The results indicated a non-linear relationship between the mean number of actions and proficiency, as well as some notable interactions between correctly answering an item, item format, response time, and response time latency. Implications for test construction and future analyses in this area are discussed.