Empirical Evaluation of Link Deletion Methods for Limiting Information Diffusion on Social Media

Abstract

Although beneficial information abounds on social media, the dissemination of harmful information such as so-called ``fake news'' has become a serious issue. Therefore, many researchers have devoted considerable effort to limiting the diffusion of harmful information. A promising approach to limiting diffusion of such information is link deletion methods in social networks. Link deletion methods have been shown to be effective in reducing the size of information diffusion cascades generated by synthetic models on a given social network. In this study, we evaluate the effectiveness of link deletion methods by using actual logs of retweet cascades, rather than by using synthetic diffusion models. Our results show that even after deleting 10\%--50\% of links from a social network, the size of cascades after link deletion is estimated to be only 50\% the original size under the optimistic estimation, which suggests that the effectiveness of the link deletion strategy for suppressing information diffusion is limited. Moreover, our results also show that there is a considerable number of cascades with many seed users, which renders link deletion methods inefficient.

Figures (12)

• Figure 1: Information flows before blocking information diffusion. Information posted by user A spreads to user A's followers, namely, users B and D. The information is reposted by user B, then the information spreads to users C and D.
• Figure 2: Information flows after blocking information diffusion. When information diffusion from user A to user B is blocked, user B cannot repost the information. As a result, user C cannot receive information from user B.
• Figure 4: Example of non-tree diffusion graph $H_t=(R_t,E_t)$ of tweet $t$. The set of users who have tweeted or retweeted tweet $t$ is $R _t = \{ 1,2,3,4,5,6,7,8\}$, and the set of links is $E _t = \{ (1,2),(1,5),(2,3),(4,5),(5,3),(3,6), (6,7),(6,8) \}$. The seed node set that does not have any incoming links is $S _t = \{ 1,4 \}$.
• Figure 5: Graph $H' _t$ after deleting links $L = \{ (1,5),(3,6) \}$ from $H _t$. In $H'_t$, node 6 has no incoming links, and therefore user 6 will not receive tweet $t$ that was originally tweeted by users 1 and 4. Moreover, users 7 and 8 who received tweet $t$ from user 6 also will not receive tweet $t$. The nodes within the reach of the seed node set $S _t = \{ 1,4 \}$ are $\{ 1,2,3,4,5 \}$, which is the set of users who will receive tweet $t$ after the link deletion.
• Figure 7: Total size of cascades after link deletion vs. the number of deleted links (non-tree diffusion graph)