[Paper Review] Incorporating Text Analysis into Evolution of Social Groups in Blogosphere

Bogdan Gliwa, Anna Zygmunt+1 more|arXiv (Cornell University)|Aug 22, 2013

Opinion Dynamics and Social Influence3 citations

TL;DR

This paper proposes a hybrid framework integrating social network analysis with text mining to study the evolution of social groups in the blogosphere. By analyzing topic distributions using LDA and measuring topic divergence, it reveals that group longevity and user migration are strongly influenced by topic coherence and thematic stability, with high topic convergence increasing the likelihood of user group joining and low divergence reducing the risk of leaving.

ABSTRACT

Data reflecting social and business relations has often form of network of connections between entities (called social network). In such network important and influential users can be identified as well as groups of strongly connected users. Finding such groups and observing their evolution becomes an increasingly important research problem. One of the significant problems is to develop method incorporating not only information about connections between entities but also information obtained from text written by the users. Method presented in this paper combine social network analysis and text mining in order to understand groups evolution.

Motivation & Objective

• To understand how topic evolution within social groups affects their longevity, size, and structural changes.
• To model user behavior—specifically joining and leaving groups—based on topic divergence between users and groups.
• To develop a method that combines social network analysis with text mining to improve prediction of group dynamics in online communities.
• To investigate the impact of group evolution events (e.g., merging, splitting) on topic popularity and thematic shifts.
• To explore the role of topic consistency in sustaining group cohesion and influencing user migration patterns.

Proposed method

• Applies Latent Dirichlet Allocation (LDA) to extract topics from blog posts and comments, modeling each group’s thematic content.
• Computes topic divergence between users and groups using Jensen-Shannon divergence to quantify thematic alignment.
• Uses Clique Percolation Method (CPM) to detect overlapping communities in dynamic social networks derived from blog interactions.
• Tracks group evolution over time by matching communities across consecutive time steps using the Jaccard index.
• Analyzes group evolution events (birth, death, merge, split, constancy, growth) and their impact on topic popularity and diversity.
• Models user migration as a probabilistic function of topic divergence, distinguishing between joiners and leavers based on prior activity and thematic alignment.

Experimental results

Research questions

• RQ1 How do topic distributions within groups influence their duration and stability over time?
• RQ2 What is the effect of group evolution events (e.g., merging, splitting) on topic popularity and thematic diversity?
• RQ3 To what extent does topic divergence between users and groups predict user migration (joining or leaving)?
• RQ4 How does thematic consistency affect the likelihood of a group surviving or dissolving?
• RQ5 Can topic-based analysis improve the prediction of group behavior in dynamic online communities?

Key findings

• Groups with high topic coherence and consistent thematic focus exhibit significantly longer lifespans, indicating that thematic stability supports group durability.
• Merging events lead to a medium rise in the popularity of a single dominant topic, suggesting that joining groups results in thematic consolidation.
• Splitting events cause a large positive change in the popularity of one topic at the expense of others, indicating thematic fragmentation and potential topic dominance post-split.
• The probability of user joining a group increases with topic convergence, peaking at 50–100% divergence, where most migrations occur due to high case volume.
• The probability of leaving a group rises with topic divergence up to 50%, then stabilizes, indicating that moderate thematic mismatch increases attrition risk.
• A significant proportion of real group joiners (about half) were inactive in the prior time slot, suggesting that topic alignment may drive re-engagement even after inactivity.