Time2Stop: Adaptive and Explainable Human-AI Loop for Smartphone Overuse Intervention

Опубликована: Май 11, 2024

Despite a rich history of investigating smartphone overuse intervention techniques, AI-based just-in-time adaptive (JITAI) methods for reduction are lacking. We develop Time2Stop, an intelligent, adaptive, and explainable JITAI system that leverages machine learning to identify optimal timings, introduces interventions with transparent AI explanations, collects user feedback establish human-AI loop adapt the model over time. conducted 8-week field experiment (N=71) evaluate effectiveness both adaptation explanation aspects Time2Stop. Our results indicate our models significantly outperform baseline on accuracy (>32.8\% relatively) receptivity (>8.0\%). In addition, incorporating explanations further enhances by 53.8\% 11.4\% receptivity, respectively. Moreover, Time2Stop reduces overuse, decreasing app visit frequency 7.0$\sim$8.9\%. subjective data also echoed these quantitative measures. Participants preferred rated highly time accuracy, effectiveness, level trust. envision work can inspire future research systems evolve users.

Язык: Английский

---

MindShift: Leveraging Large Language Models for Mental-States-Based Problematic Smartphone Use Intervention

Опубликована: Май 11, 2024

Problematic smartphone use negatively affects physical and mental health. Despite the wide range of prior research, existing persuasive techniques are not flexible enough to provide dynamic persuasion content based on users' contexts states. We first conducted a Wizard-of-Oz study (N=12) an interview (N=10) summarize states behind problematic use: boredom, stress, inertia. This informs our design four strategies: understanding, comforting, evoking, scaffolding habits. leveraged large language models (LLMs) enable automatic generation effective content. developed MindShift, novel LLM-powered intervention technique. MindShift takes in-the-moment app usage behaviors, contexts, states, goals & habits as input, generates personalized with appropriate strategies. 5-week field experiment (N=25) compare its simplified version (remove states) baseline (fixed reminder). The results show that improves acceptance rates by 4.7-22.5% reduces duration 7.4-9.8%. Moreover, users have significant drop in addiction scale scores rise self-efficacy scores. Our sheds light potential leveraging LLMs for context-aware other behavior change domains.

Язык: Английский

---

Capturing the College Experience

Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies, Год журнала: 2024, Номер 8(1), С. 1 - 37

Опубликована: Март 6, 2024

Understanding the dynamics of mental health among undergraduate students across college years is critical importance, particularly during a global pandemic. In our study, we track two cohorts first-year at Dartmouth College for four years, both on and off campus, creating longest longitudinal mobile sensing study to date. Using passive sensor data, surveys, interviews, capture changing behaviors before, during, after COVID-19 pandemic subsides. Our findings reveal pandemic's impact students' health, gender based behavioral differences, living conditions evidence persistent patterns as We observe that while some return normal, others remain elevated. Tracking over 200 from high school graduation, provides invaluable insights into behaviors, resilience in life. Conducting long-term with frequent phone OS updates poses significant challenges apps, data completeness compliance. results offer new Human-Computer Interaction researchers, educators administrators regarding life pressures. also detail public release de-identified Experience Study dataset used this paper discuss number open research questions could be studied using dataset.

Язык: Английский

---

M3BAT: Unsupervised Domain Adaptation for Multimodal Mobile Sensing with Multi-Branch Adversarial Training

Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies, Год журнала: 2024, Номер 8(2), С. 1 - 30

Опубликована: Май 13, 2024

Over the years, multimodal mobile sensing has been used extensively for inferences regarding health and well-being, behavior, context. However, a significant challenge hindering widespread deployment of such models in real-world scenarios is issue distribution shift. This phenomenon where data training set differs from real world---the environment. While explored computer vision natural language processing, while prior research briefly addresses this concern, current work primarily focuses on dealing with single modality data, as audio or accelerometer readings, consequently, there little unsupervised domain adaptation when sensor data. To address gap, we did extensive experiments adversarial neural networks (DANN) showing that they can effectively handle shifts Moreover, proposed novel improvement over DANN, called M3BAT, multi-branch training, to account multimodality during multiple branches. Through conducted two datasets, three inference tasks, 14 source-target pairs, including both regression classification, demonstrate our approach performs unseen domains. Compared directly deploying model trained source target domain, shows performance increases up 12% AUC (area under receiver operating characteristics curves) classification 0.13 MAE (mean absolute error) tasks.

Язык: Английский

---

ProGAP: Progressive Graph Neural Networks with Differential Privacy Guarantees

Опубликована: Март 4, 2024

Graph Neural Networks (GNNs) have become a popular tool for learning on graphs, but their widespread use raises privacy concerns as graph data can contain personal or sensitive information. Differentially private GNN models been recently proposed to preserve while still allowing effective over graph-structured datasets. However, achieving an ideal balance between accuracy and in GNNs remains challenging due the intrinsic structural connectivity of graphs. In this paper, we propose new differentially called ProGAP that uses progressive training scheme improve such accuracy-privacy trade-offs. Combined with aggregation perturbation technique ensure differential privacy, splits into sequence overlapping submodels are trained progressively, expanding from first submodel complete model. Specifically, each is privately aggregated node embeddings learned cached by previous submodels, leading increased expressive power compared approaches limiting incurred costs. We formally prove ensures edge-level node-level guarantees both inference stages, evaluate its performance benchmark Experimental results demonstrate achieve up 5-10% higher than existing state-of-the-art GNNs. Our code available at https://github.com/sisaman/ProGAP.

Язык: Английский

---

Learning About Social Context From Smartphone Data: Generalization Across Countries and Daily Life Moments

Опубликована: Май 11, 2024

Understanding how social situations unfold in people's daily lives is relevant to designing mobile systems that can support users their personal goals, well-being, and activities. As an alternative questionnaires, some studies have used passively collected smartphone sensor data infer context (i.e., being alone or not) with machine learning models. However, the few existing focused on specific life occasions limited geographic cohorts one two countries. This limits understanding of inference models work terms generalization everyday multiple In this paper, we a novel, large-scale, multimodal sensing dataset over 216K self-reports from 581 young adults five countries (Mongolia, Italy, Denmark, UK, Paraguay), first understand whether feasible data, then, know behavioral country-level diversity affects inferences. We found several sensors are informative context, partially personalized multi-country (trained tested all countries) country-specific within achieve similar performance above 90% AUC, do not generalize well unseen regardless proximity. These findings confirm importance better different

Язык: Английский

---

Diversity and neocolonialism in Big Data research: Avoiding extractivism while struggling with paternalism

Big Data & Society, Год журнала: 2023, Номер 10(2)

Опубликована: Июль 1, 2023

The extractive logic of Big Data-driven technology and knowledge production has raised serious concerns. While most criticism initially focused on the impacts Western societies, attention is now increasingly turning to consequences for communities in Global South. To date, debates have private-sector activities. In this article, we start from conviction that publicly funded must also be scrutinized their potential neocolonial entanglements. end, analyze dynamics collaboration an European Union-funded research project collects data developing a social platform diversity. includes pilot sites China, Denmark, United Kingdom, India, Italy, Mexico, Mongolia, Paraguay. We present experience at four field reflect project’s initial conception, our collaboration, challenges, progress, results. then different experiences comparison. conclude while succeeded finding viable strategies avoid contributing unilateral extraction as one side circle, it been infinitely more difficult break through much subtle but no less powerful mechanisms paternalism find prevalent data-driven North–South relations. These mechanisms, however, can identified other circle.

Язык: Английский

---

Inferring Mood-While-Eating with Smartphone Sensing and Community-Based Model Personalization

ACM Transactions on Computing for Healthcare, Год журнала: 2025, Номер unknown

Опубликована: Март 11, 2025

The interplay between mood and eating episodes has been extensively researched within the fields of nutrition, psychology, behavioral science, revealing a connection two. Previous studies have relied on questionnaires mobile phone self-reports to investigate relationship eating. In more recent work, sensor data utilized characterize both behavior independently, particularly in context food diaries health applications. However, current literature exhibits several limitations: lack investigation into generalization inference models trained with from various everyday life situations specific contexts like eating; an absence using explore intersection inadequate examination model personalization techniques limited label settings, common challenge (i.e., far fewer negative reports compared positive or neutral reports). this study, we examined two separate datasets different studies: i) Mexico (N \({}_{MEX}\) = 84, 1843 mood-while-eating distribution positive: 51.7%, neutral: 38.6% negative: 9.8%) 2019, ii) eight countries \({}_{MUL}\) 678, 329K reports, including 24K 83%, 14.9%, 2.2%) 2020, which contain passive smartphone sensing self-report data. Our results indicate that generic experience decline performance contexts, such as during eating, highlighting issue sub-context shifts sensing. Moreover, discovered population-level (non-personalized) hybrid (partially personalized) modeling fall short commonly used three-class task (positive, neutral, negative). Additionally, found user-level posed challenges for majority participants due insufficient labels class. To overcome these limitations, implemented novel community-based approach, building set users similar target user. findings demonstrate can be inferred accuracies 63.8% (with F1-score 62.5) MEX dataset 88.3% 85.7) MUL models, surpassing those achieved traditional methods.

Язык: Английский

---

Influence of Demographics and Personality Traits on Physiological Responses to Improve Continuous Emotion Annotation in Video Applications

Опубликована: Апрель 23, 2025

Язык: Английский

---

Promoting Prosociality via Micro-acts of Joy: A Large-Scale Well-Being Intervention Study

Опубликована: Апрель 25, 2025

Язык: Английский

---