Influence Propagation Based Influencer Detection in Online Forum

Wen Gu, Shohei Kato, Fenghui Ren, Guoxin Su, Takayuki Ito, Shinobu Hasegawa

Abstract

Identifying influential users in online forums is a critical task for community management and information spread. Unlike traditional social networks where influence is often derived from explicit follower-friend links, online forums rely on dynamic interaction structures. This paper proposes a novel framework that constructs an Online Forum Influence Propagation Network (OFIPN) and introduces the Contribution and Relevance (CR) algorithm to detect influencers based on their ability to trigger community-wide engagement.

Introduction: Beyond Static Metrics

In platforms like Twitter or Facebook, "influence" is frequently measured through static follower counts. However, online forums such as Reddit operate on a different social logic: users rarely "follow" one another. Instead, influence is built through the ebb and flow of active discussion.

Existing methods for forum influencer detection often rely on shallow statistics, such as raw post counts. These metrics fail to capture whether a user’s activity actually "activates" others to participate. The challenge lies in identifying "seed" users who can maximize the spread of information in an environment where the social graph is constructed entirely from replies and interactions rather than permanent links.

Methodology

The researchers propose a multi-stage approach to model and solve the influence maximization problem in forums.

1. The Online Forum Influence Propagation Network (OFIPN)

The first step involves transforming raw forum data—posts, replies, and likes—into a specialized graph. In this OFIPN, nodes represent users, and directed edges represent reply interactions. Unlike standard social graphs, these edges are weighted based on the frequency and intensity of interactions.

Figure 2: The Generated Online Forum Influence Propagation Network (OFIPN), illustrating weighted interaction edges between users.

2. User Metrics: Contribution and Relevance

To quantify influence, the framework utilizes two primary metrics:

  • User Contribution ($C$): A weighted sum of a user's activity (post frequency) and their community evaluation (likes received). This allows for dynamic weighting based on whether a forum values raw activity or quality-vetted content.
  • User Relevance ($R$): A measure of the strength of the connection between two specific users, calculated by the density of replies exchanged.

3. The CR (Contribution and Relevance) Algorithm

To solve the influence maximization problem, the authors developed the CR algorithm. This heuristic selects the top $k$ influencers by identifying users who possess both high personal contribution and strong relevance to their neighbors. This is paired with the URIC (User Relevance-aware Independent Cascade) model, a diffusion simulation where the probability of user A "activating" user B is proportional to their mutual relevance.

Experimental Evaluation

Datasets

The study utilized the Temporal-reddit-reply dataset, a massive collection spanning over 10 years with 8.3 million nodes. For computational feasibility, the authors extracted two subgraphs:

  • Subgraph 1: 500 nodes and 4,977 edges (default experimental setting).
  • Subgraph 2: 1,000 nodes and 12,497 edges (to test scalability).

Key Results

The authors compared the CR algorithm against four baselines: Degree (connectivity), User Contribution (activity/likes), User Relevance (tie strength), and Random selection.

Figure 5: Performance comparison of the CR algorithm against baselines, showing the total active nodes reached as the seed set size increases.
  • Superior Spread: In default simulations, the CR algorithm achieved a total of 37.424 active nodes, outperforming the User Relevance baseline by 2.74% and the standard Degree algorithm by 2.77%.
  • Robustness and Scalability: The CR algorithm consistently outperformed all baselines across six different experimental settings, including variations in user scale and contribution distributions. Even as the subgraph size doubled to 1,000 nodes, the algorithm maintained its performance lead.

Discussion and Future Work

The core "aha moment" of this research is that true forum influencers must balance personal activity with deep community ties. A user might post frequently (high contribution) but be ignored, or have a tight-knit group of friends (high relevance) but rarely post. The CR algorithm identifies the "catalyst" users who bridge these two domains.

However, the model currently has limitations:

  1. Content Blindness: The model analyzes interaction structure but ignores the sentiment or topic of posts. Integrating Natural Language Processing (NLP) is a logical next step.
  2. Heuristic Weights: Weights for posts vs. likes were manually set (0.5/0.5). Future iterations could automate weight optimization for different forum types.

By moving from static counts to dynamic propagation models, this framework provides forum moderators and researchers with a more sophisticated toolkit for understanding community health and information flow.

Summary updated

Abstract

Influential user detection is critical in supporting the human facilitator-based facilitation in the online forum. Traditional approaches to detect influential users in the online forum focus on the statistical activity information such as the number of posts. However, statistical activity information cannot fully reflect the influence that users bring to the online forum. In this paper, we propose to detect the influencers from the influence propagation perspective and focus on the influential maximization (IM) problem which aims at choosing a set of users that maximize the influence propagation from the entire social network. An online forum influence propagation network (OFIPN) is proposed to model the influence from an individual user perspective and influence propagation between users, and a heuristic algorithm that is proposed to find influential users in OFIPN. Experiments are conducted by simulations with a real-world social network. Our empirical results show the effectiveness of the proposed algorithm.
BibTeX
@article{gu2023influence,
  author = {Wen Gu and Shohei Kato and Fenghui Ren and Guoxin Su and Takayuki Ito and Shinobu Hasegawa},
  title = {{Influence Propagation Based Influencer Detection in Online Forum}},
  journal = {IEICE Trans. Information and Systems},
  year = {2023},
  doi = {10.1587/transinf.2022iip0010},
}
Figure