Exploring agent interaction patterns in the comment sections of fake and real news

TL;DR

The paper addresses how agent interactions in comment sections differ between fake and real news by examining both the topology of comment trees and the sentiment of participants on Reddit. It uses a labeled dataset and a sentiment analyzer to compare interaction patterns, revealing that fake news tends to foster more backbone/grouped structures and more negative sentiment, while real news elicits more neutral/positive responses and tighter clustering of like sentiments; sentiment distributions stabilize early in discussion growth, and early participants appear to shape later dynamics. These findings offer theoretical insight into social contagion in online networks and have practical implications for early fake/real news detection and intervention strategies. The work also highlights methodological contributions by integrating network structure with sentiment analysis and by evaluating homophily patterns against null models.

Abstract

User comments on social media have been recognized as a crucial factor in distinguishing between fake and real news, with many studies focusing on the textual content of user reactions. However, the interactions among agents in the comment sections for fake and real news have not been fully explored. In this study, we analyze a dataset comprising both fake and real news from Reddit to investigate agent interaction patterns, considering both the network structure and the sentiment of the nodes. Our findings reveal that (i) comments on fake news are more likely to form groups, (ii) compared to fake news, where users generate more negative sentiment, real news tend to elicit more neutral and positive sentiments. Additionally, nodes with similar sentiments cluster together more tightly than anticipated. From a dynamic perspective, we found that the sentiment distribution among nodes stabilizes early and remains stable over time. These findings have both theoretical and practical implications, particularly for the early detection of real and fake news within social networks.

Figures (5)

• Figure 1: Examples of comment trees for (a) fake news, (b) real news. The yellow, blue and red colors represent neutral, negative, and positive sentiments of the comment contents, respectively. The gray color indicates that the contents have been deleted from the database. Therefore, the sentiments of these comments are unknown.
• Figure 2: (a) Box plots of the distributions of $\eta_{x}^i$, where $x\in\{-1,0,1\}$. $\eta_{x}^i$ denotes the proportion of nodes with polarity $x$ in comment tree $i$. (b) Mean and variance of $\eta_{x}^i$, where $x$ represents finer-grained emotions, including joy, disgust, anger, fear, and sadness. A permutation test was conducted on these results. The p-value of less than $0.05$ indicates a significant difference in the means between the two sets.
• Figure 3: (a,b) Average fraction of neighbors with sentiment polarity $y\in\{-1,0,1\}$ pointing to a node with polarity $x\in\{-1,0,1\}$, $r_{x,y}$, for both fake and real news. (c,d) $r_{x,y}$ in the corresponding null models, where the sentiment polarities of nodes in comment trees are randomly shuffled (results averaged over $100$ realizations). The colors blue, yellow, and red represent negative, neutral, and positive sentiments, respectively, while gray corresponds to unknown sentiment. (e) One-sample t-test statistics comparing the empirical distribution to the null models.
• Figure 4: (a,b) Proportions of nodes with different sentiment polarities at each stage for fake and real news. (c,d) Proportions of nodes with different finer-grained emotions at each stage for fake and real news. In the analysis, each comment tree is divided into five stages based on its size. The first stage (S1) corresponds to the point when the tree grows to $20\%$ of its maximum size, the second stage (S2) spans from $20\%$ to $40\%$ growth, and so on.
• Figure 5: (a, c) $r_{x,y}$ for both fake and real news in user networks, where $x$ and $y$ could each denote negative, neutral, or positive sentiment. (b, d) $r_{x,y}$ in the corresponding null models, where the sentiment polarities of nodes in the user network are randomly shuffled (results averaged over $100$ realizations). The colors blue, yellow, and red represent negative, neutral, and positive sentiments, respectively.