[Paper Review] Predicting engagement in online social networks: Challenges and opportunities

Farig Sadeque, Steven Bethard|arXiv (Cornell University)|Jul 11, 2019

Complex Network Analysis Techniques27 references4 citations

TL;DR

This survey identifies key challenges and opportunities in predicting user engagement across diverse online social networks, analyzing feature engineering and machine learning techniques across platforms. It finds that engagement prediction is highly context-dependent, with graph-based features, user activity patterns, and social influence diffusion being critical, while no universal model or feature set works across all networks.

ABSTRACT

Since the introduction of social media, user participation or engagement has received little research attention. In this survey article, we establish the notion of participation in social media and main challenges that researchers may face while exploring this phenomenon. We surveyed a handful of research articles that had been done in this area, and tried to extract, analyze and summarize the techniques performed by the researchers. We classified these works based on our task definitions, and explored the machine learning models that have been used for any kind of participation prediction. We also explored the vast amount of features that have been proven useful, and classified them into categories for better understanding and ease of re-implementation. We have found that the success of a technique mostly depends on the type of the network that has been researched on, and there is no universal machine learning algorithm or feature sets that works reasonably well in all types of social media. There is a lack of attempts in implementing state-of-the-art machine learning techniques like neural networks, and the possibility of transfer learning and domain adaptation has not been explored.

Motivation & Objective

• To examine the state of research on predicting user engagement in online social networks, particularly in community-based forums and support groups.
• To identify and classify the most effective features—such as graph centrality, neighborhood properties, and influence diffusion—for engagement prediction.
• To analyze the performance of various machine learning models in different network types and highlight limitations in current approaches.
• To explore the lack of adoption of advanced techniques like deep learning and transfer learning in engagement prediction tasks.
• To provide a comprehensive overview to guide future research in user engagement modeling across heterogeneous social media platforms.

Proposed method

• Surveyed existing research on engagement prediction, categorizing studies based on task definitions and network types.
• Classified features into categories: graph-based (e.g., betweenness, degree centrality), neighborhood properties (e.g., size, churn rate in neighborhood), and structural metrics (e.g., graph density, local clustering).
• Analyzed influence diffusion models, including modified diffusion models with positive/negative influence vectors and spread factors to capture user-level churn dynamics.
• Evaluated the role of textual content, demographic data, and temporal user lifecycle patterns in engagement prediction.
• Used comparative analysis to assess the effectiveness of different models across platforms like Twitter, forums, and health support communities.
• Applied theoretical graph theory concepts such as shortest path counts and centrality measures to quantify user roles in social networks.

Experimental results

Research questions

• RQ1 What are the primary challenges in predicting user engagement in online social networks compared to traditional churn prediction in telecom?
• RQ2 Which types of features—graph-based, behavioral, or demographic—are most predictive of user engagement across different social network types?
• RQ3 How do centrality measures and neighborhood properties influence the likelihood of user disengagement or churn?
• RQ4 To what extent do influence diffusion models improve engagement prediction accuracy compared to standard models?
• RQ5 Why is there a lack of adoption of advanced machine learning techniques like neural networks and transfer learning in engagement prediction research?

Key findings

• No universal machine learning algorithm or feature set performs well across all types of social networks, as effectiveness is highly dependent on network structure and user behavior patterns.
• Graph centrality measures such as betweenness are strong predictors of future disengagement, as central users are more exposed to network churn and social disruption.
• Neighborhood size shows a strong positive correlation with user engagement, while a higher proportion of inactive users in a user’s neighborhood increases the likelihood of churn.
• Loyal communities exhibit significantly higher graph density and lower local clustering, indicating tighter-knit and more cohesive user networks.
• Influence diffusion models that account for positive and negative influence propagation outperform models that ignore interpersonal influence, improving prediction performance.
• Despite the potential, state-of-the-art techniques like deep learning and transfer learning remain underexplored in engagement prediction, representing a major research opportunity.