As students practice a specific skill in a computerized learning environment, it is helpful to provide them with a clear goal and to continuously display their progress toward that goal. We propose a formulation of this goal as “sufficient skillful effort,” contrasting it with the traditional mastery learning perspective. We present an overview of previously proposed progress measures, discussing them using formal notation in a basic setting. We also introduce a novel method based on a variable-speed progress measure that leverages a Hidden Markov Model estimate of skill. To support analysis, we describe a simulation framework for exploring progress measures and use it to examine the properties of the studied methods. Finally, we discuss practical extensions, such as accounting for typical item response times and ensuring robustness against rapid guessing. Our proposals and discussions are informed by experience with designing and applying a progress measure in a widely used learning platform.
Personalized driving assistance systems for older drivers are promising tools for decreasing the cognitive load and promoting safe driving. However, the types of assistance that are effective for various drivers with specific characteristics are unclear. Our goal is to develop personalized driving assistance systems that adaptively provide support on the basis of driver characteristics. This pilot study focuses on driving support at stop sign intersections, and the interactions between types of support and driver characteristics on driving behavior are explored. First, we collected driving behavior data and driver characteristics information from 38 participants ( M_Age = 59.3, SD_Age = 15.1, Range_Age=19-83 ). They performed driving sessions on a simulator with driving support that differed in terms of its content and the amount of information provided. To examine how effective support differs depending on driver characteristics, we built regression models to explain driving behaviors considering the interactions between driver characteristics and support types. The interaction terms improved the adjusted R^2 by 0.08 to 0.32, indicating that driver characteristics were related to the differences in the support effect. Additionally, we employed a machine learning algorithm to uncover findings about effective support. In particular, drivers with cognitive decline tended to exhibit safe driving behaviors when provided with instructive support. In contrast, neurotic drivers seemed to have trouble processing the information provided by the driving support system. These findings reveal the importance of accounting for a wide range of driver characteristics for the personalization of support.
Gamified stress management applications often employ one-size-fits-all designs that overlook individual motivational differences, limiting their potential effectiveness. This study addresses this gap through a comprehensive three-stage investigation. First, we built on previous research that systematically reviewed stress management mobile interventions from 2003 to 2024, identifying the top ten evidence-based persuasive strategies. Second, we developed prototypes that embed these strategies within a gamified mobile application, iteratively refining them based on expert evaluations. Third, we evaluated the prototypes through a large-scale online study (n = 517), assessing perceived persuasiveness and examining the influence of HEXAD gamification user types on the effectiveness of ten persuasive strategies. Our findings revealed that while all ten strategies were effective, goal setting emerged as the most persuasive, followed by self-monitoring and rewards. Notably, preferences for specific strategies varied across gamification user types. For instance, goal setting demonstrated broad appeal but was not effective for Free Spirits, whereas suggestion showed differing effectiveness for Disruptors and Socializers compared to other user types. Our research contributes to the field of gamification and human–computer interaction by providing (1) empirical evidence of how HEXAD gamification user types predict persuasive strategy effectiveness in stress management apps, (2) identification and evaluation of the ten most prevalent persuasive strategies with universal and user-specific effectiveness patterns, and (3) actionable design guidelines for both broad implementation and HEXAD-tailored personalization. This work bridges the research in gamification, human–computer interaction, and mental health by providing theoretical insights and practical tools for moving beyond generic designs towards truly user-centred, persuasive systems in gamified mental health interventions.
Prior research has highlighted the need to balance accuracy with other exploration-oriented system objectives, such as novelty, serendipity, and diversity. However, relatively little has been done to investigate how real users perceive and experience this trade-off in music recommender systems. To address this gap, we conducted a user experiment study in which participants interacted with two contrasting music discovery tools: a seed-based discovery interface designed for accuracy, and Spotify’s Discover Weekly, which prioritizes exploration. One hundred and forty-four Spotify users completed the study in which they were asked to evaluate tracks/playlists resulting from the two music discovery tools. The results indicated that Discover Weekly performed significantly better in exploration-related measures, including diversity, novelty, and serendipity, and fewer previously known tracks. However, Discover Weekly performed significantly worse in accuracy, as measured by average track rating. In addition, overall satisfaction with the playlist is significantly influenced by both the perceived diversity of the playlist and the number of known tracks. To better understand individual differences in navigating this trade-off, we examined four music preference characteristics: music involvement, music self-identity, preference for diversity, and openness to novelty. These traits, measured via a questionnaire, allowed us to investigate their roles in moderating the effectiveness of these tools. Music involvement and music self-identity were positively associated with track ratings, regardless of the tool used. Importantly, individuals who consider music central to their self-identity demonstrated greater appreciation for the recommendations provided by Discover Weekly. Structural equation modeling further showed that music self-identity moderated users’ evaluations of Discover Weekly, such that individuals with higher music self-identity reported greater satisfaction with its recommendations, partly by attenuating the negative impact of reduced familiarity on perceived accuracy. Our findings highlight the need to balance familiarity and novelty and the importance of incorporating psychological factors into personalization strategies to enhance user satisfaction.
The integration of technology offers opportunities to use persuasive systems for promoting behavior change and tackling obesity, a critical global health issue. The Persuasive Systems Design (PSD) framework offers strategies and features that can be systematically applied to digital interventions. However, it remains unclear which system characteristics and user characteristics interact in influencing health outcomes. This study investigates how user’s perceptions of PSD features (system characteristics), interact with crucial user characteristics, namely, Stages of Change (SOC), and Need for Reflection (NFR), to influence waist circumference reduction after six months of utilizing a mobile health behavior change support system. Data from 96 participants in a randomized controlled trial were analyzed after six months of mobile health behavior change support system use. Partial Least Squares Structural Equation Modeling was employed. SOC explained 22.2
Despite growing interest in monitoring cognitive states, current studies inadequately address individual differences in physiological reactions. Whereas prior works require extensive data from each individual to personalize the model, the current article explores personalization approaches operating with minimal baseline data. We propose three novel methods to personalize the model with only baseline data available for personalization. Further, we systematically compare those to an existing baseline calibration method, a non-personalized model, and a model using all available data for personalization. We conduct experiments with four open datasets with a total of 170 participants, classifying the cognitive states with a prevalent feature-based approach and a recent large time-series foundation model, MOMENT. The experiments target stress and cognitive load detection in realistic classification tasks, which require models to adapt to a new person. The best classification scores after personalizing with minimal data were around 0.7−0.9 and 0.7 balanced accuracy in binary and three-class tasks, respectively. Two of the proposed personalization methods outperformed the non-personalized model in most cases with the feature-based approach, especially in classification tasks with more than two classes, although their performance remained lower than that of the model using all data for personalization. MOMENT showed little benefit from personalization and performed comparably to the feature-based approach even with a non-personalized model. The findings provide a critical overview of the generalizability and necessity of model personalization with little data, and valuable insights into the development of personalized cognition-aware applications.
In recent times, online social networks have significantly enhanced user experiences, with social recommendation systems facilitating easier discovery of relevant information. Advanced graph neural network-based social recommendation approaches are now incorporating higher-order social relations, such as connections between friends of friends, to uncover user preferences. However, current high-order methods overlook implicit heterogeneous social connections among users and fail to account for the dynamic evolution of user interests over time. To address this issue, the paper proposes a novel heterogeneous hypergraph model for an enhanced social recommendation. Specifically, this methodology effectively manages intricate social relationships by mining various heterogeneous preferences, incorporating user–user, user–item, and item–item interactions mining represented by knowledge hypergraph-based interactions through hypergraph convolution network (HGCN). By employing HGCN, the approach aims to amplify the impact of mined heterogeneous preferences that exhibit significantly high-order user social connections, both explicit and implicit, thereby enhancing the representations by employing hypergraph convolution neural networks to provide recommendations. Further, the absence of social data for certain users is addressed by integrating implicit social connections derived from the various heterogeneous preferences and with explicit social relationships sourced from the item–item similarity matrix represented through the hypergraph-driven heterogeneous preference model (HDHP) model. Comprehensive experimentation is conducted on three real-world datasets to showcase the efficacy of the proposed HDHP model in comparison to existing state-of-the-art techniques. The proposed model shows 0.8768, 0.8876, 0.9099, 0.4554, 0.4201, and 0.4756 of recall@10, 20, 50, and NDGC@10, 20, 50, respectively. The proposed model shows an 83.95
Virtual learning environments (VLEs) have revolutionized online education by providing flexibility and accessibility while enabling the tracking of student engagement and performance. However, identifying at-risk students remains challenging due to issues such as class imbalance, dropout complexity, scalability and concept drift. To address these challenges, this research introduces a Socially Enhanced Graph with Attention and Dynamic Concept Adaptation (SEG-ADCA) framework for Early Prediction of At-Risk Students. The framework employs the Synthetic Minority Oversampling Technique (SMOTE) to effectively address class imbalance in the dataset, ensuring adequate representation of the minority class during model training and enhancing predictions for at-risk students, especially in scenarios with significantly fewer at-risk students. It incorporates a knowledge graph representation model based on temporal nodes to capture "enrolled in" relationships between students and courses, represented as a multi-dimensional matrix. This facilitates detailed analysis of time-dependent student-course interactions, enriching the feature space for prediction. The framework integrates Social Network Analysis (SNA) to capture relationships and interaction strengths among students, enhancing the understanding of peer influence, collaboration, and isolation, and providing deeper insights into social and academic dynamics essential for identifying at-risk students. The Gated Attention-Based Diffused Graph Convolutional Network (GADGCN) combines graph convolution with gated mechanisms to extract complex spatiotemporal dependencies and dynamically assign attention weights to hidden layers, improving feature relevance and boosting prediction accuracy. The Personalized Bi-Directional Gated Recurrent Model (PBGRM) incorporates a personalized behavior Embedding (PBE) layer, contextualizing critical features with personalized insights to capture individual learning patterns. The Dynamic Concept Drift Adaptation (DCDA) block addresses evolving patterns in student data by detecting and adapting to concept drift through incremental learning and feedback mechanisms, ensuring high prediction accuracy and robustness under changing conditions. Evaluated on the OULA, Junyi, Liru, and WorldUC datasets, the SEG-ADCA achieved superior performance metrics, including 98
Phishing attacks remain a significant security threat. One approach to addressing this challenge is through personalized and adaptive anti-phishing training solutions capable of tailoring learning experiences to individual needs and context. This requires cognitive models that are predictive of individual phishing responses and are amenable to analyzing and measuring the cognitive factors underlying people’s susceptibility to phishing attacks. In this paper, we study a key challenge associated with developing cognitive models of phishing decision-making grounded in instance-based learning theory (IBLT): instance engineering. We investigate the effectiveness of different approaches to designing instances using transformer-based methods for natural language representation. This work also investigates which aspects of phishing decision-making IBL models could represent and predict. We found that using representations that consider contextual meanings assigned by humans could enable cognitive agents to predict human responses to phishing emails with high accuracy. In particular, we also found that the IBL models were predictive of the responses of participants who participated in the quick and intuitive form of the decision-making process. This work underscores cognitive models’ potential to analyze differences in individual’s responses to phishing attacks, identify gaps in security awareness, and enhance anti-phishing training effectiveness.
Graph-based recommender systems have emerged as a powerful paradigm for personalized recommendations. However, their reliance on full model retraining to incorporate new users or new interactions creates scalability barriers. The task becomes infeasible in real-life recommender systems due to excessive time and resource costs involved. To address this limitation, we propose a fast and efficient method for updating graph-based recommender models without full model retraining on new data. Instead of changing all weights, we modify only small share of user representations who have new interactions. Our approach achieves a remarkable speedup of 700x over conventional model retraining approaches, drastically reducing computational overhead while maintaining the accuracy of the recommendations. Furthermore, we integrate our method into a multi-representation architecture that combines graph- and sequential-based methods to capture different user and item representations. Extensive experiments on diverse datasets demonstrate that our approach achieves state-of-the-art recommendation accuracy while maintaining the efficiency of incremental updates, outperforming existing methods in both speed and quality.
Following the period marked by the spread of the COVID-19 virus, tourists and enthusiasts flocked to museums and archaeological parks, highlighting persistent and emerging challenges concerning the enhancement of cultural heritage. This study proposes a novel framework that leverages context-aware recommender systems (CARS) to personalize the cultural experience based on the contextual conditions in which it occurs. The framework has been specifically developed and tested within the archaeological sites of Paestum and Pompeii. The evaluation process was conducted in two phases. The first phase focused on assessing the accuracy and reliability of the proposed context-aware approach, using a dataset comprising 972 user ratings collected from visitors to the Pompeii archaeological park. The second phase measured user satisfaction through a large-scale, in situ evaluation conducted directly within the archaeological parks. This phase involved nearly 2000 participants and aimed to assess the real-world usability and perceived satisfaction of the platform through direct interaction in cultural settings. The results confirm the effectiveness of the approach, with over 90% of respondents expressing a high level of satisfaction.
Learning is at the heart of every progress the human species makes. It is most effective when it considers who we are as individuals, what learning approach we prefer and what we already know to begin with. In the digital age, we strive to capture such information in the form of a digital representation—the so-called learner model—, to tailor learning-related systems to this information and build upon it to create more personalised learning experiences. Over recent years, the proliferation of diverse models across various educational applications and disciplines has made it challenging to access targeted research. In this survey, we aim to address this gap, reviewing the latest advances in learner modelling and conducting a comprehensive analysis of the existing approaches, focusing on developments from 2014 to 2023. With the help of a systematic literature review (SLR), we want to provide designers and developers of learner models with a structured overview and simplified entrance into the topic and the field of learner models. We investigate the question: What do learner models look like and how are they filled, kept up to date and used? To this end, we analyse and classify existing approaches. Our findings provide a comprehensive and structured overview of the field of learner modelling, allowing researchers to navigate and understand the diverse approaches more easily and providing developers of learner models or adaptive systems with a practical tool to access relevant information according to their needs.
Conversational recommendation systems can greatly benefit from techniques that explain the reasons behind their actions. We propose techniques that generate explanations by resorting to an auxiliary knowledge graph and an associated knowledge embedding. By exploiting the embedding plausibility score while searching a knowledge graph, we present a method that effectively generates reasons for a recommendation. We then propose a host of techniques to generate balanced reasons both for and against a recommendation, so as to enhance user trust in a conversational recommendation system. To do so, we develop a concrete implementation of Snedegar’s theory of reasons for/against. Experiments at functional, human, and application levels demonstrate that our proposals do improve the interpretability of conversational recommendations systems with controlled computational cost.
Artificial Intelligence (AI) systems are increasingly used for decision-making across domains, raising debates over the information and explanations they should provide. Most research on Explainable AI (XAI) has focused on feature-based explanations, with less attention on alternative styles. Personality traits like the Need for Cognition (NFC) can also lead to different decision-making outcomes among low and high NFC individuals. We investigated how presenting AI information (prediction, confidence, and accuracy) and different explanation styles (example-based, feature-based, rule-based, and counterfactual) affect accuracy, reliance on AI, and cognitive load in a loan application scenario. We also examined low and high NFC individuals’ differences in prioritizing XAI interface elements (loan attributes, AI information, and explanations), accuracy, and cognitive load. Our findings show that high AI confidence significantly increases reliance on AI while reducing cognitive load. Feature-based explanations did not enhance accuracy compared to other conditions. Although counterfactual explanations were less understandable, they enhanced overall accuracy, increasing reliance on AI and reducing cognitive load when AI predictions were correct. Both low and high NFC individuals prioritized explanations after loan attributes, leaving AI information as the least important. However, we found no significant differences between low and high NFC groups in accuracy or cognitive load, raising questions about the role of this specific personality trait in AI-assisted decision-making. These findings underscore the importance of user-centric personalization in XAI interfaces, where explanation styles are tailored to users’ personality traits, cognitive characteristics, and task context, with support adapted to each individual to optimize human–AI collaboration.
In this study, we develop and validate Motives of Autonomous Players (MAP) inventory. Several models on videogame motives have been published recently, but typically these models focus either on specific videogame types, on individual games, or on a particular theory on human motivation. The MAP model takes an integrative approach that considers why people play games in general. This is done by adopting an inductive bottom-up research attitude and by focusing on motives that can be argued to be broadly applicable for all kinds of videogames, ranging from casual mobile games to massively multiplayer online role-playing games. Since the MAP model is based on extensive player data that represent a great variety of player motives, the results are widely applicable in player modeling and in understanding player–game interaction at large. The initial MAP model was developed by analyzing open-ended gaming motive descriptions (N = 1,648) by a content analysis procedure. A preliminary 101-item MAP inventory was included in a UK-based survey (N = 600). A nine-factor model was identified and further validated as a 34-item version by making a confirmatory factor analysis with a USA-based survey data (N = 600). Additional analyses on construct validity were performed for investigating how motives to play videogames predict players’ game enjoyment factors that were kept analytically distinct from general motivational factors to play videogames.
In recommendation systems, the use of a user’s interaction history as sequential information can greatly improve performance. However, user interactions with preferred items are not only sparse, but also dynamically change over time, making it difficult to learn high-quality representations of user interaction item sequences. Furthermore, recommendation systems often suffer from popularity bias; namely, popular items are disproportionately recommended. To address these problems, this paper proposes a dynamic de-biasing framework based on adversarial and contrastive learning for sequential recommendations called DACRec. Firstly, to capture the changing of user preferences and enhance the temporal representation, we introduce a temporal information attention mechanism. Secondly, DACRec utilizes contrastive learning to enable efficient encoding of user representations by capturing underlying user patterns. Finally, we utilize adversarial learning to mitigate the negative impact of popularity bias on recommendation results, achieving a balance between sequential recommendation accuracy and fairness. To evaluate the effectiveness of the DACRec framework, we conduct experiments on two widely used sequential recommendation datasets Steam and Ml-1 M. The results show that there is an improvement in the three evaluation metrics of NDCG, recall, and ARP compared with multiple baselines. Particularly, the recommendation performance is better than the baseline on all-negative sampling sparse dataset, which proves that the DACRec framework is able to utilize the popularity well to improve the recommendation performance. The code is available at https://github.com/baigeshi/DACRec .
With the ever-growing amount of news, there is an increasing need for tools capable of filtering out and tailoring the content to the wants and needs of the reader. Over the last decade, researchers have noted that the inclusion of objectives such as diversity, novelty, and serendipity plays a vital role in further improving recommender systems’ perceived value and effectiveness. However, most of these studies limit themselves to a narrow definition of diversity; they only consider individual lists of recommendations, and the impact of diversity on the actual reading behavior of users is typically not examined. In this paper, we present our news recommender system that addresses this problem and aims to increase the diversity of content selected by the user. We propose a pre-filtering graph-based approach of extending the user profile to nudge him/her along a path toward unseen news topics. Results from an online experiment (N = 288) show that (1) intra-list and consumption diversity are significantly improved with negligible impact on accuracy and (2) that the use of diversity-optimized recommender systems can lead to an increase in user satisfaction.
While existing student modeling methods focus on predicting students’ knowledge states, they often overlook the underlying cognitive processes contributing to learning. In this work, we integrate cognitive processes, specifically phases of rule learning, into student modeling, drawing inspiration from cognitive science. Rule learning involves rule search, discovery, and following, providing a systematic framework for understanding how individuals acquire and apply knowledge. We conduct two studies to explore rule learning phases in a real-world learning context. Moreover, we present a two-step approach to first predict the phases of rule learning students experience during problem solving with an intelligent tutoring system and then estimate the time spent on each predicted phase. Furthermore, we identify the relationships between the time spent on specific phases of rule learning and student performance. Our findings underscore the importance of integrating cognitive processes into student modeling for more targeted interventions and personalized support.
As mobile health use is often discontinued, there is a need to improve its personalization with recommender system algorithms. This research innovatively investigates the effect of a user-based collaborative filtering and a content-based recommender algorithm for physical activity recommendation in a longitudinal between-subjects user study with objective metrics and subjective perception questions. Eighty-eight physically inactive participants used the Android app with personalized activity and tip recommendations to motivate them to move more, of which 30 participated for at least eight weeks, resulting in 1357 selected and submitted activity recommendations. Our linear mixed model analyses investigate the evolution of objective diversity, and users’ subjective perceptions of the recommendations, star rating feedback, momentary motivation, and physical activity behavior change. These analyses show that the total objective diversity of the generated recommendations was significantly larger in the collaborative group, but suggest that both algorithms performed equally well on the subjective metrics. The findings also suggest that physical activity recommenders should offer increasing diversity over time, as users in both groups preferred higher diversity as more consumptions are submitted. This study emphasizes the value of tracking the evolving diversity of recommendations and highlights the increase in both groups in perceived accuracy, fun, star ratings, and momentary motivation as more consumptions were submitted over time. As such, this research helps understanding how recommender algorithms learn users’ physical activity preferences over time and how people perceive activity recommendations, contributing to better mobile health strategies for physically inactive individuals.
Embodied virtual agents (EVAs) are beginning to be researched to improve human–computer interaction. As EVAs become increasingly integrated into various aspects of daily life, understanding how to optimize their design to foster trust and likability among users is paramount. Leveraging insights from social psychology, particularly the concept of homophily, this study investigates the impact of perceived personality traits on user perceptions of EVAs. Specifically, we explore whether aligning the personality traits of EVAs with those of users increases engagement and fosters positive interactions. Drawing on a sample of 382 participants recruited through Amazon Mechanical Turk, we assessed participants' personality traits using the Big Five Inventory—2S, while the perceived extroversion of the agent was manipulated through facial expressions and body posture. Our findings suggest that participants were able to accurately identify the perceived extroversion of the agent (p = .014), and significant results indicate a homophily effect on trust, with participants exhibiting greater trust in agents perceived as having a similar level of extroversion (p < .01). However, no significant effect on likability was detected, suggesting a more nuanced relationship between perceived personality traits and user preferences. These findings highlight the potential of leveraging homophily in designing more engaging EVAs and underscore the importance of considering user–agent compatibility in human–computer interaction.
