Multi-layer network analysis of deliberation in an online discussion platform: the case of Reddit

TL;DR

This paper uses a multi-layer network model to study deliberation in online discussion platforms, focusing on the Reddit platform, and finds that subreddits that are based on geographical regions or focus on sports have the highest levels of deliberation.

Abstract

This paper uses a multi-layer network model to study deliberation in online discussion platforms, focusing on the Reddit platform. The model comprises two layers: a discussion layer, which represents the comment-to-comment replies as a hierarchical tree, and an actor layer, which represent the actor-to-actor reply interactions. The interlayer links represent user-comment ownership. We further propose several different network metrics to characterise the level of deliberation in discussion threads, and apply the model and metrics to a large Reddit dataset containing posts from 72 subreddits focused on different topics. We compare the level of deliberation that occurs on different subreddits, finding that subreddits that are based on geographical regions or focus on sports have the highest levels of deliberation. Analysis of the actor layer reveals several features consistent across all subreddits, such as small-world characteristics and similar numbers of highly active users.

Figures (9)

• Figure 1: Initially, this two-layer network is based on Post 1, enclosed within a black dashed box. The discussion layer is structured as a radial tree, where nodes represent comments. Edges between these nodes, depicted by blue arrows, indicate reply interactions. Sink nodes are highlighted with red outlines. The actor layer, on the other hand, consists of four users contributing to Post 1, resulting in four distinct nodes. Edges between these nodes, represented as black arrows, signify one user replying to the other's comment. Meanwhile, the edges between the actor layer and the discussion layer, shown as red dashed arrows, illustrate the users responsible for these comments. The dyadic conversation is highlighted within the yellow dashed oval in the discussion layer. Subsequently, the model expand to Post 2, where the activity layer and the actor layer includes five and three additional nodes, respectively. The edges and weights within the actor layer network is also expanded accordingly.
• Figure 2: Comparison of the thread structure of two post. The maximum depth and width for these two posts are both 14 and 25, respectively. The average depth for Post 1 and Post 2 is 1.56 and 4.89, respectively. The average width for Post 1 and Post 2 is 2.78 and 10.4, respectively. The original posts are red nodes.
• Figure 3: Comparison between the 'maximum width-maximum depth' metric and the 'average width-average depth' metric. In (a), all posts within the 'AITAH' subreddit are plotted on the 'maximum width-maximum depth' coordinate system. The red lines divide the coordinate system into four quadrants, with Quadrant I containing 10% of posts across all 72 subreddits. In (b), the posts located in Quadrant I in (a) are selected and plotted on 'average width-average depth' coordinate system. The green lines divide the coordinate system into four quadrants, with Quadrant I containing 10% of posts across all 72 subreddits.
• Figure 4: Reversing the process in Fig. \ref{['plot_transfer']}. In (a), all posts within the 'AITAH' subreddit are plotted on the 'average width-average depth' coordinate system. In (b), the posts located in Quadrant I in (a) are selected and plotted on the 'maximum width-maximum depth' coordinate system.
• Figure 5: Jitter plot with overlaid distribution curves to represent the degree of deliberation across various categories. The noise is added along the y-axis. The index numbers corresponding to each subreddit can be referenced in Table \ref{['tab:subreddit_group']}.