Mining Online Social Data for Detecting Social Network Mental Disorders

Hong-Han Shuai, Chih-Ya Shen, De-Nian Yang, Yi-Feng Lan, Wang-Chien Lee, Philip S. Yu, Ming-Syan Chen

Introduction

The pervasive nature of social media has given rise to a new class of psychological challenges known as Social Network Mental Disorders (SNMDs). These include conditions such as Cyber-Relationship Addiction, Net Compulsion, and Information Overload. Traditionally, diagnosing these disorders has relied on passive methods: clinical interviews and self-reported questionnaires. However, these methods are often reactive, occurring only after significant symptoms have manifested. Researchers are now turning to data-driven approaches to enable proactive detection by analyzing the digital footprints users leave behind on platforms like Facebook and Instagram.

The Challenges of Digital Detection

Transitioning from clinical observation to automated detection involves two primary hurdles. First is the Feature Gap: server logs contain raw data such as "likes," timestamps, and check-ins, which do not directly map to psychological states like loneliness or low self-esteem. Second is the Multi-Source Problem: users often maintain different personas and activity levels across multiple platforms. A "lurker" on Facebook might be a highly active content creator on Instagram. To gain a comprehensive view of a user's mental state, researchers must integrate data from disparate sources without succumbing to the "curse of dimensionality."

The SNMDD Framework

To address these challenges, a two-phase framework called SNMDD (Social Network Mental Disorder Detection) has been developed. This framework bridges the gap between raw data and psychological insights through a structured pipeline of feature extraction and machine learning.

Figure 1: The two-phase framework for detecting Social Network Mental Disorders (SNMDs).

Phase 1: Feature Extraction

Instead of focusing on simple metrics like total time spent online—which research shows is a poor indicator of addiction—the framework extracts "proxies" for psychological behavior:

  • Parasocial Relationships: This measures the asymmetry in social interactions. A high ratio of outgoing actions (likes, comments) compared to received actions often correlates with feelings of loneliness.
  • Temporal Bursts: Utilizing Kleinberg’s burst detection algorithm, the system identifies periods of intense, irregular activity. These patterns mimic the "relapse" and "tolerance" cycles observed in clinical addiction.
  • Social Searching vs. Browsing: The model distinguishes between active searching (goal-oriented interaction) and passive browsing (scrolling), as the former is more closely linked to the brain's reward systems.
  • Online/Offline Ratio: By comparing digital posts to real-world check-ins, the system can quantify a user's preference for virtual over physical social environments.

Phase 2: SNMD-based Tensor Model (STM)

The second phase involves the SNMD-based Tensor Model (STM), which uses tensor factorization to integrate data from multiple social networks. A key innovation of the STM is the inclusion of a regularization term based on social correlation. This assumes that mental health states can "propagate" through social circles; users whose close friends exhibit signs of SNMD are statistically more likely to be at risk themselves. The final classification is performed using a Transductive Support Vector Machine (TSVM) in a semi-supervised learning environment.

Experimental Methodology

The researchers validated the framework using a combination of controlled and large-scale datasets:

  1. User Study Dataset: Data from 3,126 users recruited via Amazon Mechanical Turk. Most importantly, these users were labeled as "normal" or "potential SNMD cases" by professional psychiatrists, providing a rigorous ground truth for the model.
  2. Large-Scale Datasets: A Facebook dataset (63,000 nodes) and an Instagram dataset (2,000 users) were used to test the model's scalability and its ability to detect patterns in real-world environments.

Key Insights

The study yielded several critical insights into the nature of digital addiction:

  • Volatility Over Duration: Heavy usage does not equate to addiction. "Heavy users" typically exhibit stable, predictable patterns. In contrast, "addicted users" show volatile, bursty behavior and signs of "tolerance," where usage intensity increases over time.
  • The Multi-Source Advantage: Detecting disorders on a single platform yielded accuracies between 78% and 83%. By integrating multiple sources via the tensor model, the system could "fill in the blanks" (e.g., using public Instagram data to infer state for a private Facebook profile), significantly improving performance.
  • Structural Mental Health: The discovery of social clustering suggests that Cyber-Relationship Addiction is not just an individual issue but a structural one, where users seek validation from others who are similarly isolated in the physical world.

Results and Future Directions

The STM framework achieved a remarkable 89.7% accuracy and an AUC of 0.926 when integrating data from both Facebook and Instagram. This significantly outperformed baseline methods like simple data concatenation (75.5%). Temporal behavior and the online/offline interaction ratio emerged as the most discriminative features.

While promising, the researchers highlight the need for robust data privacy protections before such systems can be deployed. Future work aims to incorporate Natural Language Processing (NLP) and Computer Vision to analyze the actual content of posts and images, moving beyond metadata to a deeper semantic understanding of user behavior.

Summary updated

Abstract

An increasing number of social network mental disorders (SNMDs), such as Cyber-Relationship Addiction, Information Overload, and Net Compulsion, have been recently noted. Symptoms of these mental disorders are usually observed passively today, resulting in delayed clinical intervention. In this paper, we argue that mining online social behavior provides an opportunity to actively identify SNMDs at an early stage. It is challenging to detect SNMDs because the mental factors considered in standard diagnostic criteria (questionnaire) cannot be observed from online social activity logs. Our approach, new and innovative to the practice of SNMD detection, does not rely on self-revealing of those mental factors via questionnaires. Instead, we propose a machine learning framework, namely, Social Network Mental Disorder Detection (SNMDD), that exploits features extracted from social network data to accurately identify potential cases of SNMDs. We also exploit multi-source learning in SNMDD and propose a new SNMDbased Tensor Model (STM) to improve the performance. Our framework is evaluated via a user study with 3126 online social network users. We conduct a feature analysis, and also apply SNMDD on large-scale datasets and analyze the characteristics of the three SNMD types. The results show that SNMDD is promising for identifying online social network users with potential SNMDs.
BibTeX
@inproceedings{shuai2016mining,
  author = {Hong-Han Shuai and Chih-Ya Shen and De-Nian Yang and Yi-Feng Lan and Wang-Chien Lee and Philip S. Yu and Ming-Syan Chen},
  title = {{Mining Online Social Data for Detecting Social Network Mental Disorders}},
  booktitle = {The Web Conference},
  year = {2016},
  doi = {10.1145/2872427.2882996},
}
Figure