Identifying preferred routes of sharing information on social networks

TL;DR

The paper addresses whether information diffusion on social networks is random or structured, proposing two topic-sensitive preferential dynamics (global and local) that bias diffusion on an evolving underlying network. It treats information flow as biased random walks and develops two measures, the modified weighted Jaccard index and functional similarity, to detect persistent sharing routes across topics. Using real-world Farsi X political hashtags and simulations, the study shows evidence of nonrandom, topic-dependent preferred routes, with functional similarity providing a stronger match to observed data than Jaccard. The results highlight how content type shapes diffusion paths and offer quantitative tools for identifying and simulating preferential information flow on social platforms.

Abstract

The spread of information has become faster and wider than ever with the advent of social network platforms. The question raised in this study is whether information dissemination in social networks is random or follows a discernible structure. Our results from real-world hashtag data suggest that the spread of hashtags is not random and follows specific patterns. This study proposes two preferential models to explore how news spreads on social media. Specifically, we examine global and local preferential selection models and demonstrate that information dissemination aligns with these patterns. According to these two models, information flows are distributed through specific paths on networks. This suggests that new information tends to propagate along the same paths as previous news, with the specific pathways varying depending on the type of content. Finally, an examination of the propagation of political hashtags on Twitter confirms the existence of these paths that also emerge from the two preferential models.

Figures (6)

• Figure 1: One case of preferred routes in nature is riverbedspixabay_swamp, which form over time after many repetitions. The more water streams down a path, the more indented the riverbed will become. We take this idea to explain the formation of preferred routes of information sharing on social media.
• Figure 2: The weight of the undirected link A-B can render different preferences depending on the sender node, hence we are able to model a phenomenon that is inherently directed using an undirected network.
• Figure 3: Overlap of retweet imprints of two hashtags: the red (lighter color) and the black (darker color) graphs on the left represent the sharing frequency of two distinct but contextually-similar hashtags between mutual users. The solid lines on the right show the mutual links used in the spread of both hashtags, while the dotted links represent those links used only for the sharing of one hashtag. For the example case of above, the modified weighted Jaccard index is calculated as $\tilde{J}_w(A,B) = \frac{8+7+11}{8+7+11+2+2+1}\approx0.84$.
• Figure 4: Retweet imprints of a single node in two hashtags: In sharing two different hashtags, a node of interest (marked with gradient color) can act with partial similarity, calculated by Eq \ref{['eq:functional_similarity']}. The functional similarity of the node $C$ in the above example will be $S_{C}^{black,red} = \braket{p^{(black)}_C | p^{(red)}_C} \approx 0.82$. One way of measuring the overall similarity of two hashtags is to take the average of the functional similarities of all their nodes.
• Figure 5: Result of measures of the existence of preference in the dissemination of tweets.