Unveiling polarization in social networks: A matrix factorization approach

Abstract

Abstract-This paper presents unsupervised algorithms to uncover polarization in social networks (namely, Twitter) and identify polarized groups. The approach is language-agnostic and thus broadly applicable to global and multilingual media. In cases of conflict, dispute, or situations involving multiple parties with contrasting interests, opinions get divided into different camps. Previous manual inspection of tweets has shown that such situations produce distinguishable signatures on Twitter, as people take sides leading to clusters that preferentially propagate information confirming their individual cluster-specific bias. We propose a model for polarized social networks, and show that approaches based on factorizing the matrix of sources and their claims can automate the discovery of polarized clusters with no need for prior training or natural language processing. In turn, identifying such clusters offers insights into prevalent social conflicts and helps automate the generation of less biased descriptions of ongoing events. We evaluate our factorization algorithms and their results on multiple Twitter datasets involving polarization of opinions, demonstrating the efficacy of our approach. Experiments show that our method is almost always correct in identifying the polarized information from real-world twitter traces, and outperforms the baseline mechanisms by a large margin.

Quantifying and Minimizing Risk of Conflict in Social Networks

“…1. 3 Note that λ c increases with λ, λ ic decreases with λ, and λ ec first increases to reach a maximum value of 0.25 at λ = 1 after which it decreases again.…”

“…Random Walk Controversy (RWC) scores are used to quantify controversy in [15] as the difference between the properties of a random walk ending in different opinion partitions. Amin et al studied the problem of identifying and separating polarization using a matrix factorization based gradient descent algorithm [3].…”

“…However, recent work has often studied the problem in a more generic manner (independent of the topic of controversy) [21,25,30]. The identification of controversial issues has been studied using tools from sentiment analysis [7,26], as well as by relying on the structure of the social network and the distribution of opinions across it [1,3,8,9]. Besides identifying or quantifying controversy or conflict, the question of how it can be influenced has received increasing amounts of attention [17,25,28].…”

Quantifying Controversy on Social Media

“…Identifying Quantifying Content Network Choi et al [8] Popescu and Pennacchiotti [45] Mejova et al [39] Klenner et al [33] Tsytsarau et al [49] Dori-Hacohen and Allan [14] Jang et al [29] Conover et al [10] Coletto et al [9] Akoglu [2] Amin et al [3] Guerra et al [27] Morales et al [40] Garimella et al [20] inside communities created by like-minded people, who reinforce and endorse the opinions of each other. This phenomenon has been quanti ed in many recent studies [4,18,26].…”

The Age of Social Sensing

“…A particularly interesting problem is to estimate ground truth from social media outputs in the presence of polarization, a situation when two (or more) distinct camps in the social network propagate largely disjoint sets of claims that are often conflicting. Recent work addressed the challenge of detecting polarization and identifying bias of individual sources in polarized communities [10]. Experiments showed that, when accounting for such bias, state reconstruction from social observations tends to align more closely with ground truth in the physical world than when polarization is not considered.…”