SL
S.X. Li
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
3 records found
1
To support adolescents' well-being, mobile health (mHealth) applications aimed at promoting habits that improve well-being and prevent mental health problems are being increasingly developed. To increase uptake and sustained usage, advancements in personalization for these applications have been made. However, personalization is not straightforward, as it should consider the diverse and sometimes conflicting needs of stakeholders: adolescents may prioritize enjoyment and desire low time and effort, while clinical experts emphasize adherence and usefulness. Existing reinforcement learning (RL) approaches either only optimize for one of such outcomes, or collapse several objectives into a single scalar number. As a result, these methods may fail to adequately capture and address all stakeholder needs, especially when objectives conflict. This thesis explores the use of multi-objective reinforcement learning (MORL) for personalization in an mHealth application for adolescents that is aimed at promoting four different coping categories. Through literature, we identified relevant needs and concerns from adolescents, clinical experts, parents, and society, and translated these into the design of our personalization algorithm through model components and algorithmic choices. We implemented a model-based MORL algorithm based on decomposition (MORL/D) with policy iteration, using data from 317 participants from an existing study. This produces a set of Pareto optimal policies, each representing a different trade-off between the objectives. We further proposed and evaluated three metapolicies for selecting from this Pareto set: user choice, expert priority, and a combination of both. Simulation-based evaluations revealed that conflicting objectives exist and that no single-objective policy is able to improve all outcomes simultaneously, revealing inherent trade-offs across and within stakeholders. In contrast, the user choice metapolicy that provides users with three options from the Pareto set achieved consistent improvements across all outcomes compared to a non-personalized baseline, while addressing adolescents' desire for autonomy and control by allowing them to select from a set of good options. This approach enhances user experience, engagement, and diversity of the practiced coping strategies and thus may better equip adolescents with a broad set of coping skills. These results support the use of MORL and demonstrate it can be a promising approach for mHealth personalization when balancing diverse and potentially conflicting stakeholder needs.
...
To support adolescents' well-being, mobile health (mHealth) applications aimed at promoting habits that improve well-being and prevent mental health problems are being increasingly developed. To increase uptake and sustained usage, advancements in personalization for these applications have been made. However, personalization is not straightforward, as it should consider the diverse and sometimes conflicting needs of stakeholders: adolescents may prioritize enjoyment and desire low time and effort, while clinical experts emphasize adherence and usefulness. Existing reinforcement learning (RL) approaches either only optimize for one of such outcomes, or collapse several objectives into a single scalar number. As a result, these methods may fail to adequately capture and address all stakeholder needs, especially when objectives conflict. This thesis explores the use of multi-objective reinforcement learning (MORL) for personalization in an mHealth application for adolescents that is aimed at promoting four different coping categories. Through literature, we identified relevant needs and concerns from adolescents, clinical experts, parents, and society, and translated these into the design of our personalization algorithm through model components and algorithmic choices. We implemented a model-based MORL algorithm based on decomposition (MORL/D) with policy iteration, using data from 317 participants from an existing study. This produces a set of Pareto optimal policies, each representing a different trade-off between the objectives. We further proposed and evaluated three metapolicies for selecting from this Pareto set: user choice, expert priority, and a combination of both. Simulation-based evaluations revealed that conflicting objectives exist and that no single-objective policy is able to improve all outcomes simultaneously, revealing inherent trade-offs across and within stakeholders. In contrast, the user choice metapolicy that provides users with three options from the Pareto set achieved consistent improvements across all outcomes compared to a non-personalized baseline, while addressing adolescents' desire for autonomy and control by allowing them to select from a set of good options. This approach enhances user experience, engagement, and diversity of the practiced coping strategies and thus may better equip adolescents with a broad set of coping skills. These results support the use of MORL and demonstrate it can be a promising approach for mHealth personalization when balancing diverse and potentially conflicting stakeholder needs.
Student report
(2025)
-
C. Hernando De La Fuente, K.J. Trouwee, M.P. Jimenez Moreno, S.X. Li, M.A. Narkar, B.V. van Vliet, Amir Niknam, B.J.E. de Bruin
The spread of climate change mis- and disinformation poses a big threat to society; addressing this is the goal of the Joint Interdisciplinary Project (JIP) 6.1.1, in collaboration with the National Police. This report details the development of the Climate Disinformation Tracker, an open-source proof-ofconcept tool designed to trace the earliest online occurrence of climate denial narratives on Platform X and provide insightful visualizations of related tweets. The methodology, adapted from the DisTrack architecture, utilizes KeyBERT for keyword extraction and a custom scraping pipeline relying on the Nitter front-end for data retrieval, followed by mDeBERTa-v3-base-mnli-xnli for natural language inference (NLI) to classify posts as entailing, neutral, or contradictory to a user-provided claim. Validation testing demonstrated that the tool correctly identified the source tweet in 72% of claims when incorporating the synonym component, thus validating the potential of this approach for misinformation
tracking. The primary constraints identified are the dependence on non-deterministic Nitter scraping, which introduces operational instability and a 500-character query limit, and the accuracy ceiling of the alignment model. Despite these limitations, the tool validates a functional approach for empowering the public and investigative journalists with traceable context. ...
tracking. The primary constraints identified are the dependence on non-deterministic Nitter scraping, which introduces operational instability and a 500-character query limit, and the accuracy ceiling of the alignment model. Despite these limitations, the tool validates a functional approach for empowering the public and investigative journalists with traceable context. ...
The spread of climate change mis- and disinformation poses a big threat to society; addressing this is the goal of the Joint Interdisciplinary Project (JIP) 6.1.1, in collaboration with the National Police. This report details the development of the Climate Disinformation Tracker, an open-source proof-ofconcept tool designed to trace the earliest online occurrence of climate denial narratives on Platform X and provide insightful visualizations of related tweets. The methodology, adapted from the DisTrack architecture, utilizes KeyBERT for keyword extraction and a custom scraping pipeline relying on the Nitter front-end for data retrieval, followed by mDeBERTa-v3-base-mnli-xnli for natural language inference (NLI) to classify posts as entailing, neutral, or contradictory to a user-provided claim. Validation testing demonstrated that the tool correctly identified the source tweet in 72% of claims when incorporating the synonym component, thus validating the potential of this approach for misinformation
tracking. The primary constraints identified are the dependence on non-deterministic Nitter scraping, which introduces operational instability and a 500-character query limit, and the accuracy ceiling of the alignment model. Despite these limitations, the tool validates a functional approach for empowering the public and investigative journalists with traceable context.
tracking. The primary constraints identified are the dependence on non-deterministic Nitter scraping, which introduces operational instability and a 500-character query limit, and the accuracy ceiling of the alignment model. Despite these limitations, the tool validates a functional approach for empowering the public and investigative journalists with traceable context.
Smoking remains one of the largest health concerns worldwide, which is why eHealth applications with virtual coaches have been developed to assist smokers with quitting. Providing additional feedback from human coaches during such smoking cessation programs can further improve the effectiveness of the intervention. However, due to budgetary constraints and the limited availability of human coaches, it is important to make informed decisions about when someone gets human support to optimize the effectiveness. This research investigates the use of reinforcement learning (RL) to determine when to provide human feedback in quitting smoking with a virtual coach. Using data from a longitudinal study, we implemented an RL model that decides when to involve a human coach based on users' appreciation for human support and their self-efficacy, optimizing the effort that people spend on preparatory activities and their likelihood of returning to the program. Results show that the model is effective in allocating human support, increasing users' effort and return likelihood while considering the cost of human coaches. These findings support using RL to help with determining when to provide human support in smoking cessation programs.
...
Smoking remains one of the largest health concerns worldwide, which is why eHealth applications with virtual coaches have been developed to assist smokers with quitting. Providing additional feedback from human coaches during such smoking cessation programs can further improve the effectiveness of the intervention. However, due to budgetary constraints and the limited availability of human coaches, it is important to make informed decisions about when someone gets human support to optimize the effectiveness. This research investigates the use of reinforcement learning (RL) to determine when to provide human feedback in quitting smoking with a virtual coach. Using data from a longitudinal study, we implemented an RL model that decides when to involve a human coach based on users' appreciation for human support and their self-efficacy, optimizing the effort that people spend on preparatory activities and their likelihood of returning to the program. Results show that the model is effective in allocating human support, increasing users' effort and return likelihood while considering the cost of human coaches. These findings support using RL to help with determining when to provide human support in smoking cessation programs.