Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Online Social Network Data-Driven Early Detection on Short-Form Video Addiction

Fang-Yu Kuo

TL;DR

This work tackles early detection of Short-Form Video Addiction (SFVA) using online social network data. It introduces the first real SFVA dataset collected from Instagram and proposes EarlySD, an early-detection framework that combines a heterogeneous graph neural network with LLM-enhanced structural augmentation to address sparsity and missing features. The approach constructs a two-type graph with users and topics, augments edges via feature and topic similarities, and learns embeddings for SFVA classification, outperforming several baselines and ablations on multiple metrics. Together, these contributions provide a scalable, data-driven foundation for early SFVA screening with potential public-health impact on social-media platforms.

Abstract

Short-form video (SFV) has become a globally popular form of entertainment in recent years, appearing on major social media platforms. However, current research indicate that short video addiction can lead to numerous negative effects on both physical and psychological health, such as decreased attention span and reduced motivation to learn. Additionally, Short-form Video Addiction (SFVA) has been linked to other issues such as a lack of psychological support in real life, family or academic pressure, and social anxiety. Currently, the detection of SFVA typically occurs only after users experience negative consequences. Therefore, we aim to construct a short video addiction dataset based on social network behavior and design an early detection framework for SFVA. Previous mental health detection research on online social media has mostly focused on detecting depression and suicidal tendency. In this study, we propose the first early detection framework for SFVA EarlySD. We first introduce large language models (LLMs) to address the common issues of sparsity and missing data in graph datasets. Meanwhile, we categorize social network behavior data into different modalities and design a heterogeneous social network structure as the primary basis for detecting SFVA. We conduct a series of quantitative analysis on short video addicts using our self-constructed dataset, and perform extensive experiments to validate the effectiveness of our method EarlySD, using social data and heterogeneous social graphs in the detection of short video addiction.

Online Social Network Data-Driven Early Detection on Short-Form Video Addiction

TL;DR

This work tackles early detection of Short-Form Video Addiction (SFVA) using online social network data. It introduces the first real SFVA dataset collected from Instagram and proposes EarlySD, an early-detection framework that combines a heterogeneous graph neural network with LLM-enhanced structural augmentation to address sparsity and missing features. The approach constructs a two-type graph with users and topics, augments edges via feature and topic similarities, and learns embeddings for SFVA classification, outperforming several baselines and ablations on multiple metrics. Together, these contributions provide a scalable, data-driven foundation for early SFVA screening with potential public-health impact on social-media platforms.

Abstract

Short-form video (SFV) has become a globally popular form of entertainment in recent years, appearing on major social media platforms. However, current research indicate that short video addiction can lead to numerous negative effects on both physical and psychological health, such as decreased attention span and reduced motivation to learn. Additionally, Short-form Video Addiction (SFVA) has been linked to other issues such as a lack of psychological support in real life, family or academic pressure, and social anxiety. Currently, the detection of SFVA typically occurs only after users experience negative consequences. Therefore, we aim to construct a short video addiction dataset based on social network behavior and design an early detection framework for SFVA. Previous mental health detection research on online social media has mostly focused on detecting depression and suicidal tendency. In this study, we propose the first early detection framework for SFVA EarlySD. We first introduce large language models (LLMs) to address the common issues of sparsity and missing data in graph datasets. Meanwhile, we categorize social network behavior data into different modalities and design a heterogeneous social network structure as the primary basis for detecting SFVA. We conduct a series of quantitative analysis on short video addicts using our self-constructed dataset, and perform extensive experiments to validate the effectiveness of our method EarlySD, using social data and heterogeneous social graphs in the detection of short video addiction.
Paper Structure (30 sections, 6 equations, 3 figures, 8 tables)

This paper contains 30 sections, 6 equations, 3 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Mental Health Detection with Social Media Data
  4. Graph Neural Network on Social Network
  5. Large Language Model as Enhancer on Graph-Related Task
  6. Preliminary Analysis
  7. Real Data Collection
  8. Feature Extraction
  9. Quantitative Insight from demographic and social behavior feature
  10. Problem Formulation
  11. METHODOLOGY
  12. EarlySD Framework
  13. LLM-enhanced Structural Augmentation
  14. u-u Edge Augmentation
  15. u-t Edge Augmentation
  16. ...and 15 more sections

Figures (3)

  • Figure 1: Significant difference between SFVA and Non-SFVA shows age, social anxiety tendency and personality are potentially effective features.
  • Figure 2: EarlySD is comprised of two main components: (1) social network graph construction and LLM-enhanced structural augmentation and (2) heterogeneous graph neural network encode and an MLP projection head.
  • Figure 3: Expanded topic set capture additional information from user generated content