An Explainable Multimodal Framework for Adolescent Short Video Addiction Risk Prediction
This study proposes an explainable multimodal framework that integrates diverse behavioral and psychological data with a dual-attention mechanism and XGBoost to dynamically predict adolescent short-video addiction risks and their negative impacts on sleep, academics, emotions, and social adaptation.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the digital age, the way young people consume information has shifted dramatically. Short videos, characterized by their brief duration and rapid pace, have become a primary source of entertainment and social connection for adolescents. While these platforms offer instant engagement, researchers have long worried about the fine line between healthy enjoyment and harmful dependency. Traditional methods for identifying addiction have relied heavily on asking teenagers how much time they spend watching or having them fill out questionnaires about their feelings. However, these self-reported accounts often miss the subtle, shifting patterns of behavior that signal a problem is developing. They struggle to capture the dynamic nature of addiction, where a user might start with harmless viewing but gradually lose control, affecting their sleep, schoolwork, and emotional well-being. The core challenge for scientists is to move beyond simple time limits and understand the complex web of behaviors that define a risky relationship with technology.
To address this, a researcher led by Qing Bian at the Guangzhou College of Commerce has developed a new way to measure and predict the risk of short video addiction among teenagers. Instead of relying on a single number like total viewing hours, the study constructs a dynamic picture using a wide variety of data sources. The researcher gathered information on how teens use short video platforms, how often they unlock their phones, their sleep schedules, their academic performance, and their psychological states. By weaving these different strands of information together, they created a system that looks not just at what a student is doing, but at how their behavior is changing over time and how it is affecting their daily life.
The heart of this new approach is a computer model designed to act like a highly attentive observer. It does not treat all data points as equally important; instead, it learns to focus on the specific behaviors and time periods that signal danger. For instance, the system recognizes that watching videos late at night or repeatedly opening an app during study hours carries more weight than simply watching a lot of content during free time. It also pays close attention to the rhythm of a teenager's life, noticing if a pattern of behavior is escalating over a week or two. This dual focus allows the model to distinguish between a student who is simply enjoying a hobby and one whose usage is beginning to disrupt their sleep, grades, or emotional stability.
When the researcher tested this system, they found it was significantly better at identifying at-risk teenagers than previous methods. The model successfully predicted addiction risks with a high degree of accuracy, correctly flagging students who were sliding into dangerous patterns before the problem became severe. More importantly, the system did not just stop at labeling a student as "at risk." It provided a detailed explanation of why. It could pinpoint whether the primary danger was coming from sleep deprivation, a decline in academic focus, emotional distress, or difficulties in social interactions. This level of detail is crucial because it allows parents, teachers, and mental health professionals to tailor their interventions. A student whose main issue is staying up too late needs a different kind of help than a student who is using videos to escape emotional stress.
The study also explored how different timeframes affect the accuracy of these predictions. They discovered that looking at behavior over a seven-day window provided the best balance. A shorter window might react too quickly to a single bad day, while a longer window might miss a sudden, dangerous spike in risky behavior. The seven-day period aligned well with the natural rhythm of a student's week, capturing the cumulative effect of nightly usage and weekend habits without losing sensitivity to rapid changes.
Perhaps the most significant finding is that the system works by combining different types of evidence. When the researcher removed any single piece of the puzzle—such as the data on sleep patterns or psychological states—the model's ability to predict risks dropped noticeably. This confirms that addiction is not a single behavior but a complex state involving multiple parts of a teenager's life. The system effectively acts as a comprehensive health monitor, integrating digital footprints with real-world outcomes to create a clear, interpretable warning signal.
The implications of this work extend beyond simple monitoring. By generating a personalized risk profile for each individual, the framework offers a path toward early, targeted support. For a teenager showing signs of potential risk, the system might suggest a gentle reminder to manage screen time before bed. For those showing moderate risk, it could alert school counselors to provide emotional support or academic assistance. In cases of high risk, it signals the need for a coordinated effort between family and professionals to prevent further harm. The researcher emphasizes that the goal is not to punish or label students, but to provide actionable insights that help them regain control over their digital habits.
This study represents a shift from static, one-size-fits-all assessments to a dynamic, data-driven understanding of adolescent well-being. It demonstrates that by listening to the full story told by a teenager's digital and physical behaviors, we can identify problems earlier and intervene more effectively. The system does not claim to solve the issue of addiction on its own, but it provides a powerful tool for understanding the mechanics of the problem. It offers a way to see the invisible patterns that lead to harm, turning raw data into a clear guide for protecting the mental health of the next generation.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.