News Information Decoupling: An Information Signature of Catastrophes in Legacy News Media
Kristoffer L. Nielbo, Rebekah B. Baglini, Peter B. Vahlstrup, Kenneth C. Enevoldsen, Anja Bechmann, Andreas Roepstorff
NNews Information Decoupling: An InformationSignature of Catastrophes in Legacy News Media
Kristoffer L. Nielbo a,b,c , Rebekah B. Baglini c , Peter B. Vahlstrup a,b , KennethC. Enevoldsen a , Anja Bechmann b and Andreas Roepstorff c a Center for Humanities Computing Aarhus, Jens Chr. Skous Vej 4, Building 1483, 3rd floor, DK-8000 AarhusC, Denmark b DATALAB, School of Communication and Culture, Aarhus University, Helsingforsgade 14, DK-8200 AarhusN, Denmark c Interacting Minds Centre, Jens Chr. Skous Vej 4, Building 1483, 3rd floor, DK-8000 Aarhus C, Denmark
Abstract
Content alignment in news media was an observable information effect of Covid-19’s initial phase.During the first half of 2020, legacy news media became ‘corona news’ following national outbreakand crises management patterns. While news media are neither unbiased nor infallible as sources ofevents, they do provide a window into socio-cultural responses to events. In this paper, we use legacyprint media from Denmark to empirically derive the principle News Information Decoupling (NID)that functions as an information signature of culturally significant catastrophic event. Formally, NIDprovides input to change detection algorithms and points to several unsolved research problems inthe intersection of information theory and media studies.
Keywords
Newspapers, Pandemic Response, Change Detection, Adaptive Filtering “Nothing travels faster than the speed of light with the possible exception of bad news, whichobeys its own special laws.” — Douglas Adams
Introduction
As the first wave of Covid-19 virus spread across the world, content alignment of news sto-ries could be observed both within and between media sources. During December 2019 andFebruary 2020, Covid-19 news stories were, outside China, interspersed with news coverageof other events (e.g., Hong Kong protests, Iranian–American confrontation, Trump impeach-ment). As the virus spread across Europe and America, news media front pages focused almostexclusively on the pandemic, all news sections (politics, business, sports, and arts) related toCovid-19, and breaking news became corona news in continuously updated media. From theperspective of cultural dynamics, the Covid-19 pandemic provides a natural experiment thatallows us to study the effect of a global catastrophe on the the dynamics of news media’s " [email protected] (K.L. Nielbo); [email protected] (R.B. Baglini); [email protected] (P.B. Vahlstrup);[email protected] (K.C. Enevoldsen); [email protected] (A. Bechmann);[email protected] (A. Roepstorff) ~ https://knielbo.github.io/ (K.L. Nielbo) (cid:18) © a r X i v : . [ c s . C Y ] J a n igure 1: N × R slope baseline for four national newspapers that represents the left-right political spectrum.Data are sampled before Covid-19 phase 1 in Denmark initiated. information. While news media are neither unbiased nor infallible as sources of events, theydo reflect preferences, values, and desires of a wide socio-cultural and political user spectrum.As such, news media coverage of Covid-19 functions as a proxy for how cultural informationsystems respond to unexpected and dangerous events.Previous studies have shown that variation in newspapers’ word usage is sensitive to thedynamics of socio-cultural events [1, 2, 3], can detect event-driven shifts [4], and accuratelycan model effects of change in comprehensive collections of newspapers [5]. Furthermore,the co-occurrence structure of newspapers has been shown to accurately capture thematicdevelopment [6], and, when modelled dynamically, is indicative of the evolution of culturalvalues and biases [2, 7]. Adaptive smoothing and fractal analysis of word frequencies over timehave been used to identify distinct domains of newspaper content (e.g., advertisements vs.articles) [8] and to discriminate between different classes of catastrophic events that displayclass-specific fractal signatures in, among other things, word usage in newspapers [9]. Severalstudies have shown that measures of (relative) entropy can detect fundamental conceptualdifferences between distinct periods [1], concurrent normative and ideological movements [10],and even, development of ideational factors (e.g., creative expression) in temporally dependentwritings [11, 12, 13]. More specifically, a set of methodologically related studies studies haveapplied windowed relative entropy to thematic text representations to generate signals thatcapture information novelty as a reliable content difference from the past and resonance asthe degree to which future information conforms to said novelty [10, 11]. Two recent studieshave found that successful social media content show a strong association between novelty andresonance [14], and that variation in the novelty-resonance association can predict significantchange points in historical data [15].In this study we expand upon studies of novelty and resonance in cultural dynamics bymodeling change in printed news media during the initial phase of Covid-19. Specifically, wepropose the empirically derived principle of News Information Decoupling, which explains howthe information flow in legacy media responds to catastrophic events. igure 2: Novelty (upper panel), resonance (middle panel), and event-defined N × R slopes (lower panel)for center-left newspaper Politiken before and during Covid-19 phase 1. Trend lines in the upper and middlepanel are estimated using a nonlinear adaptive filter, see appendix A.
Results
The reported results are based on a sample of fours national newspapers in Denmark thatcollectively represents moderate-left (
Politiken ) and moderate-right (
Berlingske ), and left (
In-formation ) and right-wing political observation (
Ekstrabladet ). This paper focuses on moderatenewspapers during phase 1 of Covid-19, but all results have been confirmed for the full polit-ical spectrum as well as all national newspapers in Denmark (see Appendix A for details onmethods and data).To establish a baseline for novelty and resonance, we computed the pr. newspaper linearslope for resonance on novelty ( N × R ) from December 01, 2019 to February 26, 2020 (the firstcase of Covid-19 in Denmark was registered on February 27, 2020). As can be observed fromFigure 1, the slopes are remarkably similar, indicating a medium to strong association betweennovelty and resonance ( M = 0 . , SD = 0 .
06) before the national outbreak of Covid-19. Inthe normal state of affairs, novelty and resonance therefore seems to be coupled such thatnovel news items tend to resonate more than overused and repetitive items and vice versa.This general news dynamic confirms the intuition that news media, all things being equal,maintain their relevance by propagating news.The normal state of affairs was however severely disrupted by the rapid spread of Covid-19and following lockdown on March 13. For
Politiken in Figure 2 it is apparent that between thefirst case (‘Virus’) and the ‘Lockdown’ there is a non-linear decrease in novelty (upper panel)which is countered by an inverted resonance response (middle panel). This initial decoupling igure 3:
Novelty (upper panel), resonance (middle panel), and event-defined N × R slopes (lower panel)for center-right newspaper Berlingske before and during Covid-19 phase 1. of novelty and resonance does, in other words, coincide with the catastrophic Covid-19 event.While slightly less pronounced, the same decoupling patterns can be observed in
Berlingske ,see Figure 3. In fact, decoupling of novelty and resonance can be observed in all nationalbroadsheet newspapers irrespective of political observation and rare news phenomena whereall news items are concerned with one event for an extended period of time ( > Politiken was quite fast at re-establishing an almost normal state of affairs, the center-right
Berlingske showed a delayedlinear novelty response for the remainder of the first phase. This is likely to reflect the levelof support for the center-left government’s decision making regarding lockdown and opening.The center-right newspaper seems to retain the decoupling for longer as they continue to focuson the economic and societal consequences of the government’s Covid-19 policies.The N × R computed on event-defined windows, e.g., ‘ Lockdown → Opening ’, can be usedfor simple change detection technique, see Figures 2 and 3 lower panels. As phase 1 unfolds, thedecoupling is observed as loss of association between novelty and resonance. In the ‘
V irus → Lockdown ’ window (middle plot in lower panel), the slope approaches zero (for
Politiken the slope is actually negative), when the decoupling is most pronounced. The partial returnto normal can be observed in the (dis-)similarity between ‘ → W uhan ’ and ‘
Opening → ’( Politiken : 0 . → .
6, and
Berlingske : 0 . → . oncluding Remarks In conclusion we propose the empirically-derived
News Information Decoupling (NID) princi-ple, which states that in response to unexpected and dangerous temporally extended events, theordinary information dynamics of news media are (initially) decoupled such that the contentnovelty decreases as media focus monotonically on the catastrophic event, but the resonantproperty of said content increases as its continued relevance propagate throughout the newsinformation system. As such, NID behavior of the news information flow is a signature ofcatastrophes. At the level of methodology, this paper has illustrated how NID behavior oflegacy media can be identified using non-linear adaptive filtering and windowed linear fits forresonance and novelty, and function provide input for change point detection algorithms.There are multiple research paths that could contribute further to our understanding of NID.First, the predictive value of NID should be validated in terms of crisis management; secondevaluation of NID’s event scope is necessary, for instance, does NID only apply to a smallset of negative events or does it generalize to a substantially larger set of significant events(e.g., moon landing, fall of the Berlin wall); third, comparisons of information micro-behavioras a function of newspaper values, for instance, left vs. right-wing newspapers or tabloid vs.broadsheet newspapers; and finally, multilingual validation of NID.
AppendixA. Methods
Data and Normalization
The data set consists of all linguistic content (title and body text) from front pages of fourDanish national newspapers
Politiken , Berlingske , Information and
Ekstrabladet . The news-papers were sampled during December 1, 2019 to June 1 2020. Content not produced by thenewspaper, e.g., advertisements, was excluded from the sample. In order to normalize linguis-tic content, numerals and highly frequent function words were removed, and the remainingdata were lemmatized and casefolded. Subsequently, the data were represented as a bag-of-words (BoW) model using latent Dirichlet allocation in order to generate a dense low-rankrepresentation of each article. Note that with a few modifications to equations (4) and (5), theapproach works for any probabilistic or geometric vector-representation of documents. Noveltyand resonance were estimated for in windows of one week ( w = 7). Novelty and Resonance
Two related information signals were extracted from the temporally sorted BoW model:
Nov-elty as an article s ( j ) ’s reliable difference from past articles s ( j − , s ( j − , . . . , s ( j − w ) in window w : N w ( j ) = 1 w w ∑︂ d =1 J SD ( s ( j ) | s ( j − d ) ) (1)and resonance as the degree to which future articles s ( j +1) , s ( j +2) , . . . , s ( j + w ) conforms toarticle s ( j ) ’s novelty: w ( j ) = N w ( j ) − T w ( j ) (2)where T is the transience of s ( j ) : T w ( j ) = 1 w w ∑︂ d =1 J SD ( s ( j ) | s ( j + d ) ) (3)The novelty-resonance model was originally proposed in [10], but here we propose a sym-metrized and smooth version by using the Jensen–Shannon divergence ( J SD ): J SD ( s ( j ) | s ( k ) ) = 12 D ( s ( j ) | M ) + 12 D ( s ( k ) | M ) (4)with M = ( s ( j ) + s ( k ) ) and D is the Kullback-Leibler divergence: D ( s ( j ) | s ( k ) ) = K ∑︂ i =1 s ( j ) i × log s ( j ) i s ( k ) i (5) Nonlinear Adaptive Filtering
To model global trends in the novelty and resonance signals, we apply a nonlinear adaptivemulti-scale decomposition algorithm [16]. First, the signal is partitioned into overlappingsegments of length w = 2 n + 1, where neighboring segments overlap by n + 1 points. Ineach segment, the signal is fitted with the best polynomial of order M , obtained by usingthe standard least-squares regression; the fitted polynomials in overlapped regions are thencombined to yield a single global smooth trend. Denoting the fitted polynomials for the i − th and ( i + 1) − th segments by y i ( l ) and y ( i +1) ( l ), respectively, where l , l = 1 , · · · , n + 1, wedefine the fitting for the overlapped region as y ( c ) ( l ) = w y ( i ) ( l + n ) + w y ( i +1) ( l ) , l = 1 , , · · · , n + 1 (6)where w = (︁ − l − n )︁ and w = l − n can be written as (1 − d j /n ) for j = 1 ,
2, and where d j denotes the distances between the point and the centers of y ( i ) and y ( i +1) , respectively. Notethat the weights decrease linearly with the distance between the point and the center of thesegment. Such a weighting is used to ensure symmetry and effectively eliminate any jumps ordiscontinuities around the boundaries of neighboring segments. As a result, the global trendis smooth at the non-boundary points, and has the right and left derivatives at the boundary[17].The global trend thus determined can be used to maximally suppress the effect of complexnonlinear trends on the scaling analysis. The parameters of each local fit is determined bymaximizing the goodness of fit in each segment. The different polynomials in overlappedpart of each segment are combined using Equation 6 so that the global fit will be the best(smoothest) fit of the overall time series. Note that, even if M = 1 is selected, i.e., the localfits are linear, the global trend signal will still be nonlinear.Finally, in order to describe the information states before and after an events (e.g., Lockdown,Opening), we fit resonance on novelty to estimate the N × R slope β in the specific timewindows: R i = β + β N i + ϵ i , i = 1 , . . . , n. (7) . Online Resources All data are proprietary and have been collected through Infomedia’s API: https://infomedia.dk/. For inquiries regarding models and derived data, please contact [email protected]. Thesource code is available on Github: https://bit.ly/3beahFd. MOre details on NID detectioncan be found at NeiC’s NDHL website: https://bit.ly/3bfeW9C.
Acknowledgments
This paper has been supported the ”HOPE - How Democracies Cope with COVID-19”-projectfunded by The Carlsberg Foundation with grant CF20-0044 and NeiC’s Nordic Digital Human-ities Laboratory project. The authors would like to thank Infomedia for access to proprietarydata.
References [1] J. Guldi, The Measures of Modernity: The New Quantitative Metrics of Historical ChangeOver Time and Their Critical Interpretation, International Journal for History, Cul-ture and Modernity 7 (2019) 899–939. URL: https://brill.com/view/journals/hcm/7/1/article-p899 42.xml. doi: .[2] J. van Eijnatten, R. Ros, The Eurocentric Fallacy. A Digital-Historical Approach to theConcepts of ‘Modernity’, ‘Civilization’ and ‘Europe’ (1840–1990), International Jour-nal for History, Culture and Modernity 7 (2019) 686–736. URL: https://brill.com/view/journals/hcm/7/1/article-p686 33.xml. doi: .[3] J. Daems, T. D’haeninck, S. Hengchen, T. Zere, C. Verbruggen, ‘Workers of the World’ ?A Digital Approach to Classify the International Scope of Belgian Socialist Newspapers,1885–1940, Journal of European Periodical Studies 4 (2019) 99–114. URL: https://ojs.ugent.be/jeps/article/view/10187. doi: .[4] M. Kestemont, F. Karsdorp, M. D¨uring, Mining the Twentieth Century’s Historyfrom the Time Magazine Corpus, in: Proceedings of the 8th Workshop on LanguageTechnology for Cultural Heritage, Social Sciences, and Humanities (LaTeCH), Asso-ciation for Computational Linguistics, Gothenburg, Sweden, 2014, pp. 62–70. URL:http://aclweb.org/anthology/W14-0609. doi: .[5] P. Bos, H. Wijfjes, M. Piscaer, G. Voerman, Quantifying “Pillarization”: ExtractingPolitical History from Large Databases of Digitized Media Collections, Proceedings of the3rd HistoInformatics Workshop (2016) 10.[6] D. J. Newman, S. Block, Probabilistic topic decomposition of an eighteenth-centuryAmerican newspaper, Journal of the American Society for Information Science and Tech-nology 57 (2006) 753–767. URL: http://doi.wiley.com/10.1002/asi.20342. doi: .[11] J. Murdock, C. Allen, S. DeDeo, Exploration and Exploitation of Victorian Sciencein Darwin’s Reading Notebooks, arXiv preprint arXiv:1509.07175 (2015). URL: http://arxiv.org/abs/1509.07175.[12] K. L. Nielbo, M. L. Perner, C. P. Larsen, J. Nielsen, D. Laursen, Automated Compo-sitional Change Detection in Saxo Grammaticus’ Gesta Danorum, in: DHN, 2019, pp.320–332.[13] K. L. Nielbo, K. F. Baunvig, B. Liu, J. Gao, A curious case of entropic decay: Persistentcomplexity in textual cultural heritage, Digital Scholarship in the Humanities 34 (2019).URL: https://doi.org/10.1093/llc/fqy054. doi: .[14] K. Nielbo, P. Vahlstrup, A. Bechmann, Trend Reservoir Detection: Minimal Persistenceand Resonant Behavior of Trends in Social Media, Proceedings of Computational Hu-manities Research 1 (2021).[15] E. Vrangbæk, K. Nielbo, Composition and Change in De Civitate Dei: A Case Study ofComputationally Assisted Methods, Studia Patristica (2021).[16] J. Gao, J. Hu, W.-w. Tung, Facilitating Joint Chaos and Fractal Analysis of Biosignalsthrough Nonlinear Adaptive Filtering, PLoS ONE 6 (2011) e24331. URL: http://dx.plos.org/10.1371/journal.pone.0024331. doi: .[17] M. A. Riley, S. Bonnette, N. Kuznetsov, S. Wallot, J. Gao, A tutorial introduction to adap-tive fractal analysis, Frontiers in Physiology 3 (2012). URL: http://journal.frontiersin.org/article/10.3389/fphys.2012.00371/abstract. doi:10.3389/fphys.2012.00371