Abstract
The polarization of society over controversial social issues has been the subject of study in social sciences for decades (Isenberg in J Personal Soc Psychol 50(6):1141–1151, 1986, Sunstein in J Polit Philos 10(2):175–195, 2002). The widespread usage of online social networks and social media, and the tendency of people to connect and interact with like-minded individuals has only intensified the phenomenon of polarization (Bakshy et al. in Science 348(6239):1130–1132, 2015). In this paper, we consider the problem of measuring and reducing polarization of opinions in a social network. Using a standard opinion formation model (Friedkin and Johnsen in J Math Soc 15(3–4):193–206, 1990), we define the polarization index, which, given a network and the opinions of the individuals in the network, it quantifies the polarization observed in the network. Our measure captures the tendency of opinions to concentrate in network communities, creating echo-chambers. Given this numeric measure of polarization, we then consider the problem of reducing polarization in the network by convincing individuals (e.g., through education, exposure to diverse viewpoints, or incentives) to adopt a more neutral stand towards controversial issues. We formally define the ModerateInternal and ModerateExpressed problems, and we prove that both our problems are NP-hard. By exploiting the linear-algebraic characteristics of the opinion formation model we design polynomial-time algorithms for both problems. Our experiments with real-world datasets demonstrate the validity of our metric, and the efficiency and the effectiveness of our algorithms in practice.
Similar content being viewed by others
References
Adamic LA, Glance N (2005) The political blogosphere and the 2004 u.s. election: Divided they blog. In: International workshop on link discovery, LinkKDD
Akoglu L (2014) Quantifying political polarity based on bipartite opinion networks. In: International conference on weblogs and social media, ICWSM
Amelkin V, Singh AK, Bogdanov P (2015) A distance measure for the analysis of polar opinion dynamics in social networks. arXiv:1510.05058
Bakshy E, Messing S, Adamic L (2015) Exposure to ideologically diverse news and opinion on Facebook. Science 348(6239):1130–1132
Bessi A, Zollo F, Vicario MD, Puliga M, Scala A, Caldarelli G, Uzzi B, Quattrociocchi W (2016) Users polarization on Facebook and Youtube. PLoS ONE 11(8):e0159641
Bindel D, Kleinberg JM, Oren S (2015) How bad is forming your own opinion? Games Econ Behav 92:248–265
Cambria E, Poria S, Bisio F, Bajpai R, Chaturvedi I (2015) The CLSA model: a novel framework for concept-level sentiment analysis. Springer International Publishing, Cham. doi:10.1007/978-3-319-18117-2_1
Cambria E, Poria S, Bajpai R, Schuller BW (2016) SenticNet 4: A semantic resource for sentiment analysis based on conceptual primitives. In: 26th International conference on computational linguistics (COLING 2016), Proceedings of the conference: Technical Papers, Osaka, Japan, December 11–16, 2016, pp. 2666–2677
Chen T, Xu R, He Y, Xia Y, Wang X (2016) Learning user and product distributed representations using a sequence model for sentiment analysis. IEEE Comp Int Mag 11(3):34–44. doi:10.1109/MCI.2016.2572539
Conover M, Ratkiewicz J, Francisco MR, Gonçalves B, Menczer F, Flammini A (2011) Political polarization on Twitter. In: International conference on weblogs and social media ICWSM
Dandekar P, Goel A, Lee DT (2013) Biased assimilation, homophily, and the dynamics of polarization. Proc Natl Acad Sci 110(15):5791–5796
Davis G, Mallat S, Zhang Z (1994) Adaptive time-frequency decompositions with matching pursuits. Opt Eng 33(7):2183–2191
Del Vicario M, Scala A, Caldarelli G, Stanley HE, Quattrociocchi W (2017) Modeling confirmation bias and polarization. Sci Rep 7:40391. doi:10.1038/srep40391
Feige U (2003) Vertex cover is hardest to approximate on regular graphs. Technical report MCS03-15 of the Weizmann Institute
Friedkin NE, Johnsen E (1990) Social influence and opinions. J Math Soc 15(3–4):193–206
Garimella K, Morales GDF, Gionis A, Mathioudakis M (2016) Quantifying controversy in social media. In: ACM international conference on web search and data mining, WSDM, pp 33–42
Garimella VRK, Morales GDF, Gionis A, Mathioudakis M (2017) Reducing controversy by connecting opposing views. In: ACM WISDOM international conference on web search and data mining
Garrett RK (2009) Echo chambers online? Politically motivated selective exposure among internet news users1. J Comput Mediat Commun 14(2):265–285. doi:10.1111/j.1083-6101.2009.01440.x
Gionis A, Terzi E, Tsaparas P (2013) Opinion maximization in social networks. In: SIAM international conference on data mining, pp 387–395
Guerra PHC, Jr, WM, Cardie C, Kleinberg R (2013) A measure of polarization on social media networks based on community boundaries. In: International conference on weblogs and social media, ICWSM
Hager WW (1989) Updating the inverse of a matrix. SIAM Rev 31(2):221–239
Isenberg DJ (1986) Group polarization: a critical review and meta-analysis. J Personal Soc Psychol 50(6):1141–1151
Kempe D, Kleinberg J, Tardos E (2003) Maximizing the spread of influence through a social network. In: ACM SIGKDD international conference on knowledge discovery and data mining, pp 137–146
Lappas T, Crovella M, Terzi E (2012) Selecting a characteristic set of reviews. In: ACM SIGKDD international conference on knowledge discovery and data mining, pp 832–840
Lawrence P, Sergey B, Motwani R, Winograd T (1998) The pagerank citation ranking: bringing order to the web. Technical report, Stanford University
Liu B (2012) Sentiment analysis and opinion mining. Synth Lect Hum Lang Technol 5(1):1–167
Mallat S (2008) A wavelet tour of signal processing, third edition: the sparse way, 3rd edn. Academic Press, Cambridge
Munson SA, Lee SY, Resnick P (2013) Encouraging reading of diverse political viewpoints with a browser widget. In: International conference on weblogs and social media, ICWSM
Munson SA, Resnick P (2010) Presenting diverse political opinions: how and how much. In: International conference on human factors in computing systems, CHI, pp 1457–1466
Natarajan BK (1995) Sparse approximate solutions to linear systems. SIAM J Comput 24(2):227–234
Pariser E (2011) The filter bubble: what the internet is hiding from you. The Penguin Group
Poria S, Cambria E, Gelbukh A (2016) Aspect extraction for opinion mining with a deep convolutional neural network. Knowl Based Syst 108(C):42–49. doi:10.1016/j.knosys.2016.06.009
Sunstein CR (2002) The law of group polarization. J Polit Philos 10(2):175–195
Vicario MD, Scala A, Caldarelli G, Stanley HE, Quattrociocchi W (2016) Modeling confirmation bias and polarization. arXiv:1607.00022
Vydiswaran V, Zhai C, Roth D, Pirolli P (2015) Overcoming bias to learn about controversial topics. J Assoc Inf Sci Technol 66(8):1655–1672
Acknowledgements
This work was supported by the Marie Curie Reintegration Grant projects titled JMUGCS which has received research funding from the European Union, and the National Science Foundation grants: IIS 1320542, IIS 1421759, CAREER 1253393, as well as a gift from Microsoft. We would also like to thank Evaggelia Pitoura for useful comments and discussions on early drafts of the paper.
Author information
Authors and Affiliations
Corresponding author
Additional information
Responsible editor: Kurt Driessens, Dragi Kocev, Marko Robnik-Šikonja Myra Spiliopoulou.
Appendix A: Theorem 1 proof
Appendix A: Theorem 1 proof
Theorem 1 The ModerateInternal problem is NP-hard.
Proof
Our proof uses a reduction from the m-SubsetSum problem, where given a set of N positive integer numbers \(v_1,\ldots ,v_N\), a value m, and a target value b, we ask if there is a set of numbers B of size m, such that \(\sum _{v_i \in B} v_i = b\).
Given an instance of the m-SubsetSum problem, we construct an instance of ModerateInternal as follows. The graph is a star with \(N+1\) nodes: we have a central node \(u_0\), and a spoke node \(u_i\) for each integer \(v_i\). For the center of the star (node \(u_0\)) we have that \(w_{00} = t\), for an appropriately selected value of t (we will discuss this below), and \(s_0 = -1\). The weight of the edge \((u_0,u_i)\) from the center to node \(u_i\) is \(w_{0i} = v_i\), and the weight of node \(u_i\) to its internal opinion is also \(w_{ii} = v_i\). The opinion of all spoke nodes is \(s_i = 1\). We set \(k = N-m\), and we ask for a set of nodes \(T_s\), \(|T_s| = k\), such that, when setting \(s_i = 0\) for \(u_i \in T_s\) \(\pi (\mathbf {z}\mid T_z) = \Vert \mathbf {z}\Vert ^2\) is minimized.
The intuition of the proof is that the expressed opinion of the center node \(z_0\) determines \(\pi (\mathbf {z})\). The value of \(z_0\) is determined by the weight t of the internal opinion of \(u_0\), and the weights of the edges of nodes whose opinion is not set to zero. If we select t appropriately, we can guarantee that \(\Vert \mathbf {z}\Vert ^2\) is minimized when the nodes whose opinion is not set to zero sums to the value b.
Formally, assume that we have selected the set \(T_s\), \(|T_s| = k\). Assume that \(u_0 \not \in T_s\). Also let \(R = V {\setminus } T_s\cup \{u_0\}\) denote the set of spoke nodes whose opinion was not set to 0. According to the opinion formation model, the equations for the expressed opinions of the spoke nodes are as follows. For every node \(u_i \in R\), \( z_i = \frac{z_0}{2} + \frac{1}{2}. \) while for every node \(u_i \in T_s\), \( z_i = \frac{z_0}{2}. \)
We can thus write:
Recall that we want to minimize \(\pi (\mathbf {z}\mid T_s)\). To find the value of \(z_0\) that minimizes \(\pi (\mathbf {z}\mid T_s)\), we take the derivative of the expression above, we set it zero, and solve for \(z_0\). We get that the value of \(z_0\) that minimizes \(\pi (\mathbf {z})\) is:
It follows that the minimum value of \(\pi (\mathbf {z}\mid T_s)\) is
We now set the value of t such that if the set of numbers in R sums to the value of b, then \(z_0\) achieves the \(z_0^*\) value. First we compute the value of \(z_0\) as a function of t. In the following we set \(W = \sum _{i=1}^N v_i\). We have that:
Solving for \(z_0\) we get:
We want the minimum to be achieved when \(\sum _{u_i \in R} v_i = b\). Setting \(z_0 = z_0^*\) we get:
Solving for t we get:
Now, we want to prove the following. There is a set B of m numbers such that \(\sum _{v_i \in B} v_i = b\), if and only if there is a set of nodes \(T_s\) of size \(k = N-m\) such that when setting their internal opinion to zero, \(\pi (\mathbf {z}\mid T_s) < \pi ^*+\epsilon \) for some appropriate value of \(\epsilon \).
The forward direction is easy. If there exists this set B, then there is a set \(T_s\) such that when setting their opinions to zero, for the set R we have that
and therefore \(\pi (\mathbf {z}\mid T_s) = \pi ^*\).
For the backwards direction, if no such set of numbers exists, then it is not possible to find a set of nodes \(T_s\) such the nodes in R give \(z_0 = \frac{K-N}{N+4}\) that minimizes \(\pi (\mathbf {z}\mid T_s)\). Therefore, there must be an \(\epsilon \) such that \(\pi (\mathbf {z}\mid T_s) \ge \pi ^*+ \epsilon \).
To set \(\epsilon \) note that for any \(z_0 \ne z_0^*\)
where the inequality follows from the fact that the values \(v_1,\ldots ,v_N, b\) are integers and their difference is at least one. Now, let \(\mathbf {z}^*\) be the vector with \(z_0^*\) that achieves the minimum value \(\pi ^*\). For any other \(\mathbf {z}\) we have
So it suffices to set \(\epsilon < \frac{(k+4)^2}{4(N+4)(W+b)^2}\).
Finally, in our computations so far we have assumed that our set \(T_s\) does not contain node \(u_0\). This is not a restrictive assumption. Consider a solution \(T_s\), where \(u_0 \in T_s\), and \(s_0 = 0\). Then, since \(s_0\) is the only negative opinion value in our instance, it follows that \(z_0 \ge 0\), and for any node \(u_i\in R\) we have that \( z_i = \frac{1}{2} z_0 + \frac{1}{2} \ge \frac{1}{2} \). There are \(N+1-k\) nodes in R. Therefore,
Note that \(\pi ^*= (N-k)(k+4)/4(N+4) \le (N-k)/4\), since \(k \le N\). Therefore, \(\pi (\mathbf {z}) \ge 2\pi ^*+1/4\). Selecting \(\epsilon < \pi ^*+ \frac{1}{4}\) guarantees that \(\pi (\mathbf {z}| T_s) > \pi ^*+ \epsilon \). Thus, if there is a set \(T_s\) such that \(\pi (\mathbf {z}| T_s)\) is minimized, it cannot contain \(u_0\). \(\square \)
Rights and permissions
About this article
Cite this article
Matakos, A., Terzi, E. & Tsaparas, P. Measuring and moderating opinion polarization in social networks. Data Min Knowl Disc 31, 1480–1505 (2017). https://doi.org/10.1007/s10618-017-0527-9
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10618-017-0527-9