On Building Online Visualization Maps for News Data Streams by Means of Mathematical Optimization

On Building Online Visualization Maps for News Data Streams by Means of Mathematical Optimization Emilio Carrizosa (PhD)1, Vanesa Guerrero (PhD) †2, Daniel Hardt (PhD)3, and Dolores Romero Morales (PhD)3 1Instituto de Matem´aticas de la Universidad de Sevilla (IMUS), Seville, Spain [email protected] 2Departamento de Estad´ıstica, Universidad Carlos III de Madrid, Getafe, Spain [email protected] 3Copenhagen Business School, Frederiksberg, Denmark {dh.digi, drm.eco}@cbs.dk †Corresponding Author Abstract In this paper we develop a novel online framework to visualize news data over a time horizon. First, we perform a Natural Language Processing analysis, where the words are extracted, and their attributes, namely the importance and the relatedness, are calculated. Second, we present a Mathematical Optimization model for the visualization problem and a numerical optimization approach. The model represents the words using circles, the time-varying area of which displays the importance of the words in each time period. Word location in the visualization region is guided by three criteria, namely, the accurate representation of semantic relatedness, the spread of the words in the visualization region to improve the quality of the visualization, and the visual stability over the time horizon. Our approach is flexible, allowing the user to interact with the display, as well as incremental and scalable. We show results for three case studies using data from Danish news sources. Keywords: Online Visualization; News Data Streams; Word Importance; Relatedness; Visual Stability Visualizing News Data Streams 1 Introduction With the growth of Big Data, new business opportunities arise for gaining insight from data streams [2,21,29,55], and Information Visualization has played a crucial role in facilitating this task, [12,15,26,27,42,59]. However, this can be very challenging, not just because of the size of these data streams [1,6,31,50], but also because of structural complexities that make the analysis difficult [8,9,22,34]. This is the case for news data [16,19,23,35,36,44,45,49,57], where there is a virtually unlimited supply of data available, containing insights that are crucial for business and society [13]. In this paper, we show how these insights can be unlocked from news data by addressing its structural complexity. News data can be considered as a sequence of words, which can be visualized using methods such as word clouds. Word clouds aim to represent the frequency or importance of words in texts by balancing several aesthetic criteria [61,62], including avoiding overlapping and reducing the empty spaces. However, standard word clouds do not manage to convey two important features of the structure of news data. The first is temporal – it is important to know how news topics develop over time. The second is semantic – certain words and topics are closely related to other words or topics. In this paper we use Mathematical Optimization techniques [7,10] to build a novel online visualization map that makes it possible to produce visualizations of news data as it develops over time. This online visualization in turn depends on Natural Language Processing for uncovering the semantic structure in the news data, by computing the importance and the relatedness of the words that occur. In this way a kind of dynamic word cloud is produced, which displays the importance as well as the relatedness of words, and shows how they develop over time. Although there have been attempts in the literature to incorporate the temporal [20] and the semantic [4,24,66] features into word clouds, as far as the authors are aware, this is the first paper in which these are optimized 1 Visualizing News Data Streams simultaneously. The approach proposed here has important advantages. First, by being online, and therefore incremental, we do not require the recalculation of the past display whenever new data comes in, [36]. Second, our approach is scalable since the optimization problems solved in each iteration are computationally cheap. Third, the modeling is flexible, allowing the user to interact with the display, for example, by defining new criteria to be optimized, constraints on the layout of the display, and different visualization regions over time. Finally, our work employs word embeddings [41,51,52,53,56], a newly developed method that provides a much more nuanced and precise model of semantic structures. We consider a set Nof words and Ttime periods. Our aim is to build an online visualization map, i.e., a collection of Tvisualization maps, where the words in Nare represented in each time period. We have two attributes associated with the words, the so-called importance and the relatedness matrix, which are both dynamic. For each word iand each period t,ωi,t represents the importance of word iin time period t. For each pair of words iand jand each period t,δij,t represents the semantic relatedness between words iand jin period t. In each period, we depict words as circles representing accurately the importance ωi,t. The position of the circles is defined by three criteria. The first criterion takes care that, in each time period, the distance between the circles resembles the relatedness δij,t attached to the words, as in MultiDimensional Scaling [37,38,60]. The second criterion aims at spreading the circles across the visualization region to improve the aesthetics [62] and ensure the readability of the online visualization map. The third criterion ensures a smooth transition between the visualization maps, thus ensuring visual stability, [18]. Finally, the approach is online, such that the plots for the first T0−1 periods are already available, and we need to construct the plots in periods T0, . . . , T, where new words may appear, using the criteria above. With these criteria, our ordered time sequence of visualizations behaves as a kind of animated conceptual space, helping to show how words gain or recede in importance 2 Visualizing News Data Streams over time. The remainder of the paper is structured as follows. In Section 2, we review the related literature. In Section 3, we describe the way the importance and the relatedness are calculated in this paper. In Section 4, we describe the Mathematical Optimization approach used to build the online visualization map. In Section 5, we illustrate the usefulness of our tool, by applying it to three case studies from Danish news. Some concluding remarks and plans for future research are given in Section 6. 2 Literature review In this paper, we describe an online visualization framework based on Mathematical Optimization, which can extract insights from news data streams as they develop over time. Mathematical Optimization is typically used to represent dissimilarities (the complement of relatedness), see e.g., [5,28,46,47] and references therein. However, we are not aware of any similar approaches to extracting these insights from news data streams. The framework involves several key elements: •Importance: term importance must be computed, both to decide which terms to include, and their relative size •Relatedness: the semantic relatedness of terms must be calculated – this provides a basis for deciding where to place terms in the visualization region •Visual Stability: this is made possible by the computation of semantic relatedness, together with a Mathematical Optimization model which ensures that related terms appear close to each other. As a side effect of this, term position will remain stable in a temporal sequence of visualizations. This stability over time is crucial to providing the insights over time from 3 Visualizing News Data Streams a news data stream. In this section, we will review related work which addresses certain key elements of this approach, although none of them put them together in the way described here. Word Clouds have become a widespread tool for visualizing the important terms or concepts in texts. Wordle [33,62] is a widely used tool for this purpose. There have been many works to improve word clouds as data exploratory tool, see [24] and references therein. In [64], EdWordle is proposed to allow the user to edit the word cloud (e.g., adding/deleting, dragging, resizing words), while preserving its neighborhood structure. After editing, the words are rearranged based on rigid body dynamics, [65]. In [14], this consistency of the neighborhood structure is applied to dynamic data, where the shape defining the visualization region of the cloud may vary over time too. In [11], the authors develop a methodology to coordinate word clouds. They introduce the concept of a word storm, i.e., a collection of word clouds, each of them associated with a different text. Since word storms are used to compare documents, the authors argue that the same word should have a similar location across different word clouds. One of the approaches to build word storms is based on Mathematical Optimization. The objective function has three goals, namely, texts that are similar are represented by similar word clouds, the frequencies are represented accurately, and the aesthetics of the display (measured by compactness and avoidance of overlapping). Normally word clouds provide information along one dimension, namely, the frequency or importance of the term, which is represented by the size of the term in the display. Most of the effort is put into the aesthetics of the word cloud. The position of a term in the display is essentially random; it conveys no information. In the present approach, the position of the term conveys important information – namely, semantically related terms appear closer to each other, and unrelated terms are more distant. Some recent work describes word clouds that attempt to incorporate semantic relatedness. For 4 Visualizing News Data Streams example, Cui et al. [20] use semantic relatedness to visualize temporal document content evolution. The authors propose building a dynamic word cloud and a trend chart to highlight significant changes in content over time. They use a stepwise procedure to build the dynamic word cloud. First, MultiDimensional Scaling (MDS) is executed on the set of all words across all periods to choose the initial points where the words will be anchored. The authors define three possible dissimilarities to be used by MDS, one of them being based on semantic relatedness using feature vectors as in [58]. Second, a word cloud is built for each period, starting with the points corresponding to words appearing in that period, and, subsequently, the aesthetics of the display is improved with the so-called force-directed algorithm. As the authors themselves note, their approach does not allow for user interaction. Barth et al. [4] propose three approaches for building semantics-preserving word clouds, i.e., word clouds trying to represent accurately both word frequencies as well as relatedness between words. The first one starts with an MDS solution built on the rescaled problem, where both the frequencies and the dissimilarities are scaled by the same constant. The word cloud is inflated iteratively and the overlapping is removed using the procedure in [20]. The two other approaches are built around the graph defined by the dissimilarities. The second one builds a collection of stars, i.e., trees of depth one. Then, an MDS layout is built for the stars, while the force-directed algorithm is called to improve the aesthetics of the display as above. The third approach works similarly, but in this case the authors extract cycles/paths from the dissimilarity graph. Barth et al. [3] is a follow-up work to [4]. The authors provide a graph representation of the problem, where the vertices are the words and the weights on the arcs are the relatedness between them, they then show that the problem is NP-Complete, and propose approximation algorithms for specific structures. Further approaches in terms of text visualizations and analysis are surveyed in [43]. 5 Visualizing News Data Streams 3 Natural Language Processing analysis Our aim is to produce an online visualization map, which provides semantic insights into a stream of news data. This requires Natural Language Processing analysis from two perspectives: Importance and Relatedness. 3.1 Importance Importance involves extracting the most interesting words appearing in the news data stream, and providing a numerical rating of the importance of the word. In other words, we want to know whether the word indicates an important topic in the current news data stream. Computing semantic importance of words is a well-studied problem, and we follow well-established techniques, with some modifications to address issues related to our application, in which we wish to extract insights from a topical news data stream. The frequency of word occurrences provides a basic, yet flawed, measure of semantic importance. An obvious flaw concerns extremely common words that typically express particular linguistic functions, such as prepositions (in, on, of), articles (a, an, the) or conjunctions (and, or). Despite their frequency, such words do not indicate semantically important topics. It is typical to define a stopword list consisting of such words. The problem of stopwords indicates a more general problem – frequency simply does not correlate with semantic importance. A standard way of addressing this is to use a different metric: Term Frequency Inverse Document Frequency (tfidf). The idea of this metric is to reward words for appearing frequently in the particular text of interest, but to penalize words that generally tend to appear frequently across many texts. This penalty is captured by the inverse document frequency (idf) [48], which is idf =log(D/d) 6 Visualizing News Data Streams where Dis the total number of documents, and dis the number of documents where the term occurred. Thus, idf is smaller for terms that occur in many documents. The tfidf value is computed with the following formula tfidf =tf ∗idf where tf is simply the number of times the term appears in the current document – in our case the selected news data stream. In the current study, we have selected several topical news data streams, by selecting all articles in a certain period mentioning a certain key term, such as Internet,Terror Attack or Immigration. So in these cases, the document is considered to be the currently selected news stream, and we compute term frequency as all the occurrences of a term within that news data stream. To compute the inverse document frequency, idf, we selected a large collection of proceedings from the European Parliament (Europarl) [32]. The Europarl collection is a standard resource in Natural Language Processing, because it provides a large reference for standard language use for all the major European languages. It is thus excellent for our purpose, which is to measure the typical frequency of words across many texts. The importance of a given term i, for a given time period t, is given by: ωi,t =tfi,t ∗idfi where tfi,t is computed for the news text of interest in time period t, and idfiis computed in the general background text, namely the European Parliament proceedings. 3.2 Relatedness It is not enough to compute the relative importance of terms in a news stream. Some terms are closely related to other terms. For example, in the news data stream on the keyword Immigration, the terms muslim,scarf, and immigrant background are closely related. It is widely acknowledged that cooccurence data can provide some valuable information about 7 Visualizing News Data Streams words that only feature in one period out of the four periods, such as ‘Diatel’ (1995), a Danish videotex online service accessible through telephone lines, and ‘Telia’ (1996), a telephone company in Denmark. Even though the dissimilarity between words is time-independent, it is affected by words entering/leaving set N(t) from one period to the next. Nevertheless, our plots preserve these dissimilarities reasonably well. From the dendrogram in Figure 1, we can see words ‘Telia’, ‘IBM’, ‘Apple’, ‘Microsoft’, ‘Diatel’, ‘PBS’, ‘Tele Danmark’ and ‘UNI-C’ clustering together. In the visualization map, these words are placed on the right side of the plot and close to each other. This is especially true for ‘Apple’ and ‘Microsoft’, which are present in all 4 periods. For the Danish keyword Terrorangreb (Terror Attack), Table 2contains the N= 127 words and their translation into English, while their dendrogram can be found in Figure 6. The visualization map can be found in Figures 7-21. The words ‘terrorangreb’ and ‘Al-Qaeda’ feature prominently in each period, and they are located close to each other on the left-top corner of the visualization region. Examples of words that feature in fewer periods are ‘tegninger’ (drawings) and ‘profeten’ (the Prophet), that appear for the first time in 2005 and reappear in 2015. From the dendrogram in Figure 6, we can see words ‘PET’, ‘PST’, ‘Pentagon’, ‘FBI’, ‘CIA’ and ‘NSA’ clustering together. In the visualization map, these words are placed around the anti-diagonal of the visualization region and close to each other. This is especially true for ‘PET’ (Danish Security and Intelligence Service) and ‘CIA’, which are present in almost all periods. For the Danish keyword Indvandring (Immigration), Table 3contains the N= 201 words and their translation into English, while their dendrogram can be found in Figure 22. The visualization map can be found in Figures 23-43. The keyword Immigration shares several words with the previous one (Terror Attack), including the names of several Danish and Swedish political parties. The political parties are clustered around each other in the dendrogram, ‘Ny Alliance’, ‘Sverigedemokraterne’, ‘Kristeligt Folkeparti’, ‘Fremskridtspartiet’, ‘Dansk Folkeparti’ (also known 14 Visualizing News Data Streams as DF), ‘SF’ and ‘Venstre’. In the visualization map, these parties appear close to each other on the right hand side of the visualization region. 6 Conclusions In this paper we have developed a novel and flexible online framework for the visualization of news data streams. We have combined techniques from Natural Language Processing and Mathematical Optimization to extract the most relevant words associated with a given keyword, compute their importance and their relatedness, and build an online visualization map. In this map, the words have been represented by circles, the area of which indicate the importance of words accurately, and that are placed in the visualization region following three criteria, namely, resemblance of the distances between circles to the dissimilarities, spread of the circles across the visualization region and stability of the position of the circles across the time horizon. Our Mathematical Optimization approach can incorporate constraints defined by the user, other criteria to guide the optimization, other glyphs to represent the words, and a dynamic visualization region that changes over time. We have presented three case studies, using data from Danish news sources, and based on these studies we argue that the tool provides a level of insight beyond that of previously described work. There are several interesting lines for future research. First, the modeling approach in this paper is flexible and allows for desirable functionalities of data exploratory tools, such as the comparison of topics [17]. This boils down to modeling visual stability across topics, similar to the way in which we have modelled the visual stability across time, ensuring that words appear in a similar location across different topics. Second, the approach proposed in this paper can be extended to cope with other types of data streams that exhibit both temporal and semantic structure. For example, product reviews are similar to news data in possessing temporal and semantic structure. The techniques described in this paper could therefore be useful in review mining to gain insights in 15 Visualizing News Data Streams customer experience [63]. Third, we plan to explore more sophisticated linguistic structures in our analysis, which currently is restricted to individual, frequently occurring words or short phrases. In future work we will provide a more systematic treatment of multiword phrases. Fourth, in the current work the importance of a given word is based solely on the occurrences of that word. A more sophisticated treatment might involve clustering of similar words, so that what is displayed is a label selected to represent the importance of a group of similar words. Acknowledgement. This research is funded in part by Projects MTM2015-65915-R (Spain), P11- FQM-7603 and FQM-329 (Andaluc´ıa), all with EU ERD Funds. References [1] C. Albrecht-Buehler, B. Watson, and D.A. Shamma. Visualizing live text streams using motion and temporal pooling. IEEE Computer Graphics and Applications, 25(3):52–59, 2005. [2] B. Baesens. Analytics in a Big Data World: The Essential Guide to Data Science and its Applications. Wiley and SAS Business Series. Wiley, 2014. [3] L. Barth, S.I. Fabrikant, S.G. Kobourov, A. Lubiw, M. Nllenburg, Y. Okamoto, S. Pupyrev, C. Squarcella, T. Ueckerdt, and A. Wolff. Semantic word cloud representations: Hardness and approximation algorithms. In A. Pardo and A. Viola, editors, LATIN 2014: Theoretical Informatics, volume 8392 of Lecture Notes in Computer Science, pages 514–525. Springer Berlin Heidelberg, 2014. [4] L. Barth, S.G. Kobourov, and S. Pupyrev. Experimental comparison of semantic word clouds. In J. Gudmundsson and J. Katajainen, editors, Experimental Algorithms, volume 8504 of Lecture Notes in Computer Science, pages 247–258. Springer International Publishing, 2014. 16 Visualizing News Data Streams [5] A. Boytsov, F. Fouquet, T. Hartmann, and Y. LeTraon. Visualizing and Exploring Dynamic High-Dimensional Datasets with LION-tSNE. arXiv preprint arXiv:1708.04983, 2017. [6] N. Cao, Y.-R. Lin, X. Sun, D. Lazer, S. Liu, and H. Qu. Whisper: Tracing the spatiotemporal process of information diffusion in real time. IEEE Transactions on Visualization and Computer Graphics, 18(12):2649–2658, 2012. [7] E. Carrizosa, V. Guerrero, and D. Romero Morales. Visualization of complex dynamic datasets by means of mathematical optimization. Technical report, IMUS, Sevilla, Spain, 2017. [8] E. Carrizosa, V. Guerrero, and D. Romero Morales. Visualizing proportions and dissimilarities by space-filling maps: a large neighborhood search approach. Computers & Operations Research, 78:369–380, 2017. [9] E. Carrizosa, V. Guerrero, and D. Romero Morales. On mathematical optimization for the visualization of frequencies and adjacencies as rectangular maps. European Journal of Operational Research, 265:290–302, 2018. [10] E. Carrizosa, V. Guerrero, and D. Romero Morales. Visualizing data as objects by DC (difference of convex) optimization. Mathematical Programming, 169:119–140, 2018. [11] Q. Castell`a and C. Sutton. Word storms: Multiples of word clouds for visual comparison of documents. In Proceedings of the 23rd International Conference on World Wide Web, WWW’14, pages 665–676. ACM, 2014. [12] C.P. Chen and C.-Y. Zhang. Data-intensive applications, challenges, techniques and technologies: A survey on big data. Information Sciences, 275:314–347, 2014. [13] H. Chen, R.H.L. Chiang, and V.C. Storey. Business intelligence and analytics: From big data to big impact. MIS Quarterly, 36(4):1165–1188, 2012. 17 Visualizing News Data Streams [14] M.-T. Chi, S.-S. Lin, S.-Y. Chen, C.-H. Lin, and T.-Y. Lee. Morphable word clouds for timevarying text data visualization. IEEE Transactions on Visualization and Computer Graphics, 21(12):1415–1426, 2015. [15] J. Choo and H. Park. Customizing computational methods for visual analytics with big data. IEEE Computer Graphics and Applications, 33(4):22–28, 2013. [16] C. Collins, F.B. Vi´egas, and M. Wattenberg. Parallel tag clouds to explore and analyze faceted text corpora. In IEEE Symposium on Visual Analytics Science and Technology, 2009. VAST 2009, pages 91–98, 2009. [17] G. Coppersmith and E. Kelly. Dynamic wordclouds and vennclouds for exploratory data analysis. In Proceedings of the Workshop on Interactive Language Learning, Visualization, and Interfaces, pages 22–29, 2014. [18] W. Cui, S. Liu, Z. Wu, and H. Wei. How hierarchical topics evolve in large text corpora. IEEE Transactions on Visualization and Computer Graphics, 20(12):2281–2290, 2014. [19] W. Cui, H. Qu, H. Zhou, W. Zhang, and S. Skiena. Watch the story unfold with textwheel: Visualization of large-scale news streams. ACM Transactions on Intelligent Systems and Technology, 3(2):1–17, 2012. [20] W. Cui, Y. Wu, S. Liu, F. Wei, M.X. Zhou, and H. Qu. Context preserving dynamic word cloud visualization. IEEE Computer Graphics and Applications, 30(6):42–53, 2010. [21] T.H. Davenport. Big Data @ Work. Harvard Business Review Press, 2014. [22] S. Fortunato. Community detection in graphs. Physics Reports, 486(3):75–174, 2010. [23] T. Gao, J.R. Hullman, E. Adar, B. Hecht, and N. Diakopoulos. NewsViews: an automated 18 Visualizing News Data Streams pipeline for creating custom geovisualizations for news. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, pages 3005–3014. ACM, 2014. [24] E. Gomez-Nieto, W. Casaca, D. Motta, I. Hartmann, G. Taubin, and L.G. Nonato. Dealing with multiple requirements in geometric arrangements. IEEE Transactions on Visualization and Computer Graphics, 22(3):1223–1235, 2016. [25] Z.S. Harris. Distributional structure. Word, 10(2-3):146–162, 1954. [26] J. Heer, M. Bostock, and V. Ogievetsky. A tour through the visualization zoo. Communications of the ACM, 53:59–67, 2010. [27] J. Heer and B. Shneiderman. Interactive dynamics for visual analysis. Queue, 10(2):30–55, 2012. [28] G.E. Hinton and S.T. Roweis. Stochastic neighbor embedding. In Advances in Neural Information Processing Systems, pages 857–864, 2003. [29] A.H. Huang, R. Lehavy, A.Y. Zang, and R. Zheng. Analyst information discovery and interpretation roles: A topic modeling approach. Forthcoming in Management Science, 2017. [30] Infomedia. www.infomedia.dk, 2017. [31] A. Jakaitiene, M. Sangiovanni, M.R. Guarracino, and P.M. Pardalos. Multidimensional Scaling for Genomic Data, pages 129–139. Springer International Publishing, Cham, 2016. [32] P. Koehn. Europarl: A parallel corpus for statistical machine translation. In MT Summit, volume 5, pages 79–86, 2005. [33] K. Koh, B. Lee, B. Kim, and J. Seo. ManiWordle: Providing Flexible Control over Wordle. IEEE Transactions on Visualization and Computer Graphics, 16(6):1190–1197, 2010. 19 Visualizing News Data Streams [34] M. van Kreveld and B. Speckmann. On rectangular cartograms. Computational Geometry, 37(3):175–187, 2007. [35] M. Krstaji´c, E. Bertini, and D. Keim. Cloudlines: Compact display of event episodes in multiple time-series. IEEE Transactions on Visualization and Computer Graphics, 17(12):2432– 2439, 2011. [36] M. Krstaji´c, M. Najm-Araghi, F. Mansmann, and D.A. Keim. Story Tracker: Incremental visual text analytics of news story development. Information Visualization, 12(3-4):308–323, 2013. [37] J.B. Kruskal. Multidimensional scaling by optimizing goodness of fit to a nonmetric hypothesis. Psychometrika, 29(1):1–27, 1964. [38] J.B. Kruskal and M. Wish. Multidimensional Scaling, volume 11. Sage, 1978. [39] H.A. Le Thi. An efficient algorithm for globally minimizing a quadratic function under convex quadratic constraints. Mathematical Programming, 87:401–426, 2000. [40] O. Levy, Y. Goldberg, and I. Dagan. Improving distributional similarity with lessons learned from word embeddings. Transactions of the Association for Computational Linguistics, 3:211– 225, 2015. [41] S. Liu, P.-T. Bremer, J.J. Thiagarajan, V. Srikumar, B. Wang, Y. Livnat, and V. Pascucci. Visual exploration of semantic relationships in neural word embeddings. IEEE Transactions on Visualization and Computer Graphics, 24(1):553–562, 2018. [42] S. Liu, W. Cui, Y. Wu, and M. Liu. A survey on information visualization: recent advances and challenges. The Visual Computer, 30(12):1373–1393, 2014. 20 Visualizing News Data Streams [43] S. Liu, X. Wang, C. Collins, W. Dou, F. Ouyang, M. El-Assady, L. Jiang, and D. Keim. Bridging text visualization and mining: A task-driven survey. Forthcoming in IEEE Transactions on Visualization and Computer Graphics, 2018. [44] Y. Lu, M. Steptoe, S. Burke, H. Wang, J.-Y. Tsai, H. Davulcu, D. Montgomery, S.R. Corman, and R. Maciejewski. Exploring evolving media discourse through event cueing. IEEE Transactions on Visualization and Computer Graphics, 22(1):220–229, 2016. [45] D. Luo, J. Yang, M. Krstaji´c, W. Ribarsky, and D. Keim. EventRiver: Visually exploring text collections with temporal references. IEEE Transactions on Visualization and Computer Graphics, 18(1):93–105, 2012. [46] L. vander Maaten and G. Hinton. Visualizing data using t-SNE. Journal of Machine Learning Research, 9:2579–2605, 2008. [47] L. vander Maaten and G. Hinton. Visualizing non-metric similarities in multiple maps. Machine Learning, 87(1):33–55, 2012. [48] C.D. Manning, P. Raghavan, and H. Sch¨utze. Introduction to Information Retrieval. Cambridge University Press, Cambridge, 2008. [49] A. Marcus, M.S. Bernstein, O. Badar, D.R. Karger, S. Madden, and R.C. Miller. TwitInfo: aggregating and visualizing microblogs for event exploration. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, pages 227–236. ACM, 2011. [50] D. Martens and F. Provost. Explaining data-driven document classifications. MIS Quarterly, 38(1):73–99, 2014. [51] T. Mikolov, K. Chen, G. Corrado, and J. Dean. Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781, 2013. 21 Visualizing News Data Streams [52] T. Mikolov, I. Sutskever, K. Chen, G. Corrado, and J. Dean. Distributed representations of words and phrases and their compositionality. In NIPS’13 Proceedings of the 26th International Conference on Neural Information Processing Systems, pages 3111–3119, 2013. [53] T. Mikolov, W.-t. Yih, and G. Zweig. Linguistic regularities in continuous space word representations. In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 746–751, 2013. [54] J. Pennington, R. Socher, and C.D. Manning. Glove: Global vectors for word representation. In EMNLP, volume 14, pages 1532–1543, 2014. [55] F. Provost and T. Fawcett. Data Science for Business: What you need to know about data mining and data-analytic thinking. O’Reilly Media, Inc., 2013. [56] R. ˇ Reh˚uˇrek and P. Sojka. Software Framework for Topic Modelling with Large Corpora. In Proceedings of the LREC 2010 Workshop on New Challenges for NLP Frameworks, pages 45–50, Valletta, Malta, 2010. ELRA. [57] S. Rose, S. Butner, W. Cowley, M. Gregory, and J. Walker. Describing story evolution from dynamic information streams. In IEEE Symposium on Visual Analytics Science and Technology, 2009. VAST 2009., pages 99–106, 2009. [58] H. Sch¨utze. Automatic word sense discrimination. Computational Linguistics, 24(1):97–123, 1998. [59] J. Thomas and P.C. Wong. Visual Analytics. IEEE Computer Graphics and Applications, 24(5):20–21, 2004. [60] W.S. Torgerson. Theory and Methods of Scaling. Wiley, 1958. 22 Visualizing News Data Streams [61] F.B. Vi´egas and M. Wattenberg. Tag clouds and the case for vernacular visualization. Interactions, 15(4):49–52, 2008. [62] F.B. Vi´egas, M. Wattenberg, and J. Feinberg. Participatory visualization with Wordle. IEEE Transactions on Visualization and Computer Graphics, 15(6):1137–1144, 2009. [63] F. Villarroel Ordenes, B. Theodoulidis, J. Burton, T. Gruber, and M. Zaki. Analyzing customer experience feedback using text mining: A linguistics-based approach. Journal of Service Research, 17(3):278–295, 2014. [64] Y. Wang, X. Chu, C. Bao, L. Zhu, O. Deussen, B. Chen, and M. Sedlmair. EdWordle: Consistency-preserving Word Cloud Editing. IEEE Transactions on Visualization and Computer Graphics, 24(1):647–656, 2018. [65] A. Witkin. Physically based modeling: principles and practice constrained dynamics. Computer Graphics, 1997. [66] Y. Wu, T. Provan, F. Wei, S. Liu, and K.-L. Ma. Semantic-preserving word clouds by seam carving. Computer Graphics Forum, 30(3):741–750, 2011. 23 Visualizing News Data Streams Internet 1997 adressen Apple browser Dankort danskere harddisk IBM Internet_Explorer Jytte_Hilden KØBENHAVN MB Microsoft modem Netscape PBS PC RB Tele_Danmark Windows Århus Figure 5: Visualization map for Internet in 1997 30 Visualizing News Data Streams Table 2: Danish words and their translation into English for the Terror Attack case study Danish word Translation Danish word Translation Danish word Translation AC AC HK HK republikanerne Republicans aftale appointment imam imam retsordfører spokesman aften evening Irakkrigen Iraq War Reuters Reuters Afghanistan Afghanistan Islamisk Stat Islamic State Riyadh Riyadh al-Awlaki al-Awlaki israelske Israeli Romney Romney al-Qaeda al-Qaeda Jens Stoltenberg Jens Stoltenberg SAS SAS al-Shabaab Al-Shabaab jihadister jihadists Saddam Saddam al-Zarqawi al-Zarqawi John McCain John McCain Said Mansour Said Mansour Alm Alm Kabul Kabul saudiske Saudi Al Qaida Al Qaida Katrina Katrina selvmordsbomber suicide bomber Amir Amir kenyanske Kenyan SF SF AP AP Kerry Kerry somaliske Somali Bagdad Baghdad Krystalgade Krystalgade Sotji Sochi Bali Bali Kurt Westergaard Kurt Westergaard stammeomr˚aderne tribal areas Beredskabsstyrelsen Emergency Management Københavns Lufthavn Copenhagen Airport synagogen synagogue Beslan Beslan Lars Lars Taleban Taliban BH BH LH LH tegninger drawings bin bin lufthavn airport terrorangreb terrorist attacks bombemænd bomb men Manhattan Manhattan terrorbomberne bombings Boston Boston masseødelæggelsesv˚aben WMD terrormistænkte terror suspects Breivik Breivik Maximus Maximus Twitter Twitter Bush Bush Mette Frederiksen Mette Frederiksen uighurer Uighurs Charlie Hebdo Charlie Hebdo miltbrand anthrax Utøya Utøya Chicago Chicago Mombasa Mombasa Villy Søvndal Villy Søvndal CIA CIA Moore Moore Volgograd Volgograd Clarke Clarke Morten Storm Morten Storm World Trade Center World Trade Center danskere Danes moske mosque Yemen Yemen Dansk Folkeparti Danish People’s Party Muhammed Muhammed Det Hvide Hus The White House Mumbai Mumbai DF DF Musharraf Musharraf droneangreb drone attacks muslimer Muslims Edward Snowden Edward Snowden New Orleans New Orleans efterfølgende subsequent norske Norwegian efterretningstjenester intelligence services NSA NSA eftektivt effectively nyhedsbureauet news agency efterretningstjeneste intelligence service Nyrup Nyrup el-Hussein el-Hussein Obama Obama el-Sheikh el-Sheikh offentlige public Facebook Facebook Omar El-Hussein Omar El-Hussein FBI FBI Osama Osama flykaprere hijackers Oslo Oslo Fogh Fogh pakistanske Pakistani forbundspoliti Federal police Pentagon Pentagon gasanlægget gas plant PET PET Giuliani Giuliani PFA PFA Ground Zero Ground Zero profeten Prophet Hamas Hamas præsidentkandidat presidential candidate Headley Headley PST PST Helle Thorning-Schmidt Helle Thorning-Schmidt radikaliserede radicalized Hizbollah Hezbollah Rana Rana 31 Visualizing News Data Streams 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Dendrogram for Terror Attack Lars LH AC HK Alm PFA SAS el-Sheikh Katrina bin nyhedsbureauet AP Reuters gasanlægget New_Orleans Boston Chicago Mombasa Bali Manhattan stammeområderne Afghanistan Yemen Riyadh Bagdad Kabul Beslan Sotji Volgograd aftale lufthavn moske Ground_Zero Utøya terrorbomberne Oslo synagogen aften Beredskabsstyrelsen Københavns_Lufthavn Villy_Søvndal Dansk_Folkeparti DF SF Helle_Thorning-Schmidt Mette_Frederiksen retsordfører Breivik Fogh Nyrup John_McCain republikanerne præsidentkandidat Romney Det_Hvide_Hus Musharraf Bush Obama Giuliani Clarke Moore Maximus Rana effektivt forbundspoliti efterretningstjenester efterretningstjeneste Jens_Stoltenberg Kerry offentlige pakistanske somaliske norske kenyanske israelske saudiske PET PST Pentagon FBI CIA NSA efterfølgende Facebook Twitter Saddam al-Zarqawi Edward_Snowden Morten_Storm Irakkrigen masseødelæggelsesvåben Al_Qaida al-Shabaab Hamas Taleban Hizbollah al-Qaeda Islamisk_Stat droneangreb terrorangreb selvmordsbomber BH Omar_El-Hussein Krystalgade Mumbai World_Trade_Center imam Said_Mansour terrormistænkte miltbrand Amir Headley Osama al-Awlaki el-Hussein Muhammed profeten tegninger Charlie_Hebdo Kurt_Westergaard jihadister uighurer bombemænd flykaprere radikaliserede danskere muslimer Figure 6: Hierarchical clustering for dissimilarities in the Terror Attack case study 32 Visualizing News Data Streams Terror Attack 2001 ● ● ● ● ● ● ● ● ● Afghanistan al−Qaeda AP bin CIA danskere Dansk_Folkeparti FBI Fogh Islamisk_Stat israelske Kabul Manhattan miltbrand muslimer nyhedsbureauet Nyrup Osama Pentagon PET PFA SAS SF Taleban terrorangreb World_Trade_Center Figure 7: Visualization map for Terror Attack in 2001 33 Visualizing News Data Streams Terror Attack 2002 ● ● ●● ● ● ● al−Qaeda Alm Al_Qaida AP Bali Beredskabsstyrelsen bin Bush CIA danskere Det_Hvide_Hus FBI flykaprere forbundspoliti Ground_Zero Københavns_Lufthavn Manhattan Mombasa nyhedsbureauet Osama PET SAS terrorangrebWorld_Trade_Center Yemen Figure 8: Visualization map for Terror Attack in 2002 34 Visualizing News Data Streams Terror Attack 2003 ● ● ● ● ● ● ●● al−Qaeda Al_Qaida AP Bagdad Bali bin Bush CIA Det_Hvide_Hus Hamas Irakkrigen masseødelæggelsesvåben Osama PET PFA Riyadh Saddam Said_Mansour saudiske selvmordsbomber terrorangreb terrormistænkte World_Trade_Center Figure 9: Visualization map for Terror Attack in 2003 35 Visualizing News Data Streams Terror Attack 2004 ● ● ●● ● ● al−Qaeda al−Zarqawi Al_Qaida AP Bagdad Beslan bin Bush CIA Clarke Det_Hvide_Hus Irakkrigen Kerry masseødelæggelsesvåben Moore nyhedsbureauet Osama PET republikanerne saudiske terrorangreb World_Trade_Center Figure 10: Visualization map for Terror Attack in 2004 36 Visualizing News Data Streams Terror Attack 2005 ● ● ●● ● ● ● ●● ● ● al−Qaeda Al_Qaida AP Bali bin bombemænd Bush CIA danskere Det_Hvide_Hus el−Sheikh Fogh imam Irakkrigen Islamisk_Stat Katrina moske muslimer New_Orleans Osama PET selvmordsbomber terrorangreb terrorbomberne terrormistænkte Figure 11: Visualization map for Terror Attack in 2005 37 Visualizing News Data Streams Terror Attack 2006 ● ● ● ● ● AC al−Qaeda Al_Qaida AP bin CIA danskere Fogh Hamas imam Islamisk_Stat John_McCain moske Muhammed muslimer Osama PET profeten republikanerne SAS tegninger terrorangreb terrormistænkte World_Trade_Center Figure 12: Visualization map for Terror Attack in 2006 38 Visualizing News Data Streams Terror Attack 2007 aftale aften al−Qaeda Al_Qaida AP bin CIA efterfølgende efterretningstjenester effektivt Giuliani Islamisk_Stat Lars LH lufthavn Maximus moske Musharraf offentlige Osama PET republikanerne Taleban terrorangreb Figure 13: Visualization map for Terror Attack in 2007 39 Visualizing News Data Streams Terror Attack 2014 aften al−Qaeda Amir AP BH CIA efterfølgende efterretningstjenester Hamas Islamisk_Stat lufthavn norske NSA nyhedsbureauet Obama offentlige PET PST Reuters Sotji terrorangreb uighurer Figure 20: Visualization map for Terror Attack in 2014 46 Visualizing News Data Streams Terror Attack 2015 al−Qaeda AP Charlie_Hebdo danskere Dansk_Folkeparti DF el−Hussein Facebook Helle_Thorning−Schmidt Islamisk_Stat jihadister Krystalgade Mette_Frederiksen Muhammed offentlige Omar_El−Hussein PET profeten radikaliserede retsordfører synagogen tegninger terrorangreb Figure 21: Visualization map for Terror Attack in 2015 47 Visualizing News Data Streams Table 3: Danish words and their translation into English for the Immigration case study Danish word Translation Danish word Translation Danish word Translation Danish word Translation Aarhus Aarhus Frank Jensen Frank Jensen Københavns Universitet University of Copenhagen Romney Romney Abu Laban Abu Laban Frederik Frederik Lars Hedegaard Lars Hedegaard Roseng˚ard Roseng˚ard Ali Ali fremmede foreign Lars Løkke Rasmussen Lars Løkke Rasmussen sammenhængskraft cohesion andengenerationsindvandrere second generation immigrant Fremskridtspartiet Progress Party Leo Leo Sarkozy Sarkozy Anne Knudsen Anne Knudsen Fryd Joy Leyla Leyla SF SF Arizona Arizona Fædregruppen Fathers Group LO LO sgu damn arj arj Gellerup Gellerup Louise Gade Louise Gade sjællænder Sealander bandekrigen gang war ghettoer ghettos Lubna Elahi Lubna Elahi skuespiller actor Basim Basim Glistrup Glistrup Madsen Madsen skyderier shootings beduinerne Bedouins Gogh Gogh Majid Majid Sleiman Sleiman Berlingske Tidende Berlingske Tidende Grønland Greenland Malmø Malmo Socialdemokratiet Social Democratic Party Bertel Haarder Bertel Haarder Gyldent Daggry Gyldent Daggry Mamadou Mamadou socialr˚adgiver social worker Birte Weiss Birte Weiss halalhippier halal hippies Manu Sareen Manu Sareen somaliere Somalis Bispehaven Bispehaven Halle Halle Marianne Jelved Marianne Jelved stationen station Bjørn Nørgaard Bjørn Nørgaard Hamid Rahmati Hamid Rahmati Marwan Marwan Strauss-Kahn Strauss-Kahn blog blog Hans Lassen Hans Lassen Mazhar Hussain Mazhar Hussain stuepige maid Bl˚ag˚ards Plads Bl˚ag˚ards Plads Hasan Hasan McCain McCain Suleebo Suleebo bogen the book Hells Angels Hells Angels menighed congregation Sverigedemokraterne Sweden Democrats Brian Brian Henrik Henrik mentor mentor Søren Pind Søren Pind Bycyklen City Bike herberg hostelry MFP MFP taxa taxi bydelsr˚adet town council hvidløg garlic Mir Mir tegninger drawings bæveren beaver H¨useyin Arac H¨useyin Arac Mogens Camre Mogens Camre Theo Theo Carlos Carlos ideer ideas moskeer mosques Thorkild Simonsen Thorkild Simonsen Charlie Hebdo Charlie Hebdo ikke-vestlige non-Western Muhammed Muhammed Thorvaldsen Thorvaldsen Chopin Chopin imamer imams museum museum Tingbjerg Tingbjerg Cirkeline Cirkeline IND-sam IND-SAM muslimer Muslims torvet square cyklerne bikes indvandrerbaggrund immigrant background Mustafa Mustafa tosprogede bilingual Danmarks Statistik Statistics Denmark Inger Støjberg Inger Støjberg m˚arhunde raccoon dogs tvangsægteskaber forced marriages danskere Danes Instr Director Nakskov Nakskov tyrkere Turks Dansk Folkeparti Danish People?s Party integrationsminister Minister of integration Ny Alliance Ny Alliance tørklæde scarf danskundervisning danish lessons Ishøj Ishøj Nyrup Nyrup udlændingepolitik immigration policy DF DF islam Islam nævningene jury uighurer Uighurs Dieudonn´e Dieudonn´e Islamisk Stat Islamic State Nørrebro Nørrebro Ulla Dahlerup Ulla Dahlerup digte poems Jamal Jamal Obama Obama ulve wolves dingoen dingo Jeppe Jeppe Odense Odense urfuglen grouse drabsforsøg attempted murder Josef Josef OL Olympic Games vampyr vampire drenge boys jyder Jutlanders opgangen entryway Vangsgaard Vangsgaard DSK DSK jyske Native Jutland ORG ORG velfærdsturisme welfare tourism dørmænd doormen Jønke Jønke pakistanere Pakistanis Venstre Venstre ECRI ECRI Karen Jespersen Karen Jespersen PET PET Vesterbro Vesterbro Egtvedpigen Egtvedpigen Kassem Kassem PhD PhD vildsvin boar Ekstra Bladet Ekstra Bladet Kenneth Kenneth Pia Kjærsgaard Pia Kjærsgaard Villy Søvndal Villy Søvndal elg moose Khader Khader Pim Fortuyn Pim Fortuyn Vollsmose Vollsmose Esbjerg Esbjerg Khaled Ramadan Khaled Ramadan POEM POEM Wallait Khan Wallait Khan Eske Eske kontanthjælp cash assistance profeten Prophet Wilders Wilders Facebook Facebook Krasnik Krasnik Rabin Rabin Yahya Hassan Yahya Hassan familiesammenførte reunited family Krim Crimea RB RB ˚ Arhus Aarhus filmen film Kristeligt Folkeparti Kristeligt Folkeparti Reich Reich ˚ Arhusianske Native Aarhus flygel Grand piano Kristian Thulesen Dahl Kristian Thulesen Dahl Republikanerne Republicans ¨ Ozlem ¨ Ozlem Fogh Fogh kronik feature article rockere rockers Folketinget the Danish parliament Kurt Kurt romaer Roma 48 Visualizing News Data Streams 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Dendrogram for Immigration Fogh Nyrup Gyldent_Daggry Folketinget Ny_Alliance Sverigedemokraterne Kristeligt_Folkeparti Fremskridtspartiet Venstre Dansk_Folkeparti Socialdemokratiet Mogens_Camre Hüseyin_Arac Lubna_Elahi Wallait_Khan Marianne_Jelved Kristian_Thulesen_Dahl Pia_Kjærsgaard Villy_Søvndal DF SF Lars_Løkke_Rasmussen Bertel_Haarder Inger_Støjberg Frank_Jensen Søren_Pind Karen_Jespersen Birte_Weiss Thorkild_Simonsen Louise_Gade Manu_Sareen integrationsminister Özlem menighed sammenhængskraft udlændingepolitik halalhippier ideer digte tegninger bandekrigen Hells_Angels drabsforsøg stuepige Gogh Theo Jønke DSK Strauss-Kahn nævningene MFP PET Frederik Leo Jeppe Kenneth Brian Henrik Kurt Bycyklen dingoen Khaled_Ramadan Bjørn_Nørgaard bogen filmen ORG Wilders Muhammed profeten Charlie_Hebdo Lars_Hedegaard Yahya_Hassan blog kronik Facebook Ulla_Dahlerup Berlingske_Tidende Ekstra_Bladet fremmede tvangsægteskaber velfærdsturisme islam Islamisk_Stat ghettoer moskeer imamer muslimer mårhunde ulve vildsvin OL bæveren Grønland dørmænd skyderier uighurer rockere andengenerationsindvandrere drenge danskere jyder romaer tyrkere pakistanere somaliere indvandrerbaggrund tosprogede familiesammenførte danskundervisning kontanthjælp jyske århusianske Blågårds_Plads torvet Nørrebro Vesterbro Malmø Rosengård Bispehaven Gellerup Nakskov Tingbjerg Vollsmose Odense Aarhus Århus Esbjerg Ishøj flygel herberg museum hvidløg arj Instr tørklæde taxa opgangen stationen cyklerne Egtvedpigen urfuglen Hamid_Rahmati Sleiman Suleebo Fryd Mir Chopin sgu vampyr elg sjællænder Krim Pim_Fortuyn Rabin Carlos Mamadou Halle Josef Reich Eske Thorvaldsen Leyla Majid Hasan Ali Jamal Mustafa Abu_Laban Kassem Basim Dieudonné Marwan Vangsgaard mentor skuespiller socialrådgiver Københavns_Universitet ph Cirkeline Glistrup Krasnik Madsen Khader RB Arizona Romney McCain Republikanerne Obama Sarkozy beduinerne Danmarks_Statistik ikke-vestlige Mazhar_Hussain ECRI Anne_Knudsen Hans_Lassen LO IND-sam POEM bydelsrådet Fædregruppen Figure 22: Hierarchical clustering for dissimilarities in the Immigration case study 49 Visualizing News Data Streams Immigration 1995 ● ● beduinerne Bycyklen cyklerne danskere herberg hvidløg ideer IND−sam indvandrerbaggrund Kassem Københavns_Universitet Lubna_Elahi Mamadou Nakskov nævningene Odense OL opgangen Rabin RB romaer SF torvet tyrkere Vangsgaard Vesterbro Århus Figure 23: Visualization map for Immigration in 1995 50 Visualizing News Data Streams Immigration 1996 Ali Blågårds_Plads Brian bydelsrådet danskere dørmænd Folketinget Halle indvandrerbaggrund jyder jyske Kristeligt_Folkeparti Leo Lubna_Elahi Mazhar_Hussain menighed Nørrebro Odense pakistanere RB SF sgu sjællænder Århus Figure 24: Visualization map for Immigration in 1996 51 Visualizing News Data Streams Immigration 1997 andengenerationsindvandrere Birte_Weiss Blågårds_Plads danskere Dansk_Folkeparti danskundervisning Ekstra_Bladet familiesammenførte Folketinget fremmede Fremskridtspartiet indvandrerbaggrund Ishøj Khader Lubna_Elahi Nørrebro Odense Pia_Kjærsgaard RB SF somaliere taxa Thorkild_Simonsen Thorvaldsen urfuglen Venstre Århus Figure 25: Visualization map for Immigration in 1997 52 Visualizing News Data Streams Immigration 1998 Anne_Knudsen bæveren Cirkeline danskere Dansk_Folkeparti DF filmen Folketinget fremmede Fædregruppen indvandrerbaggrund integrationsminister Khader Københavns_Universitet Mogens_Camre Nyrup Nørrebro Odense Pia_Kjærsgaard RB SF Thorkild_Simonsen tørklæde vildsvin Århus Figure 26: Visualization map for Immigration in 1998 53 Visualizing News Data Streams Immigration 1999 ● ● andengenerationsindvandrere Berlingske_Tidende Bjørn_Nørgaard bæveren danskere Dansk_Folkeparti DF elg Folketinget Frederik Fædregruppen Glistrup indvandrerbaggrund Ishøj Københavns_Universitet Nyrup Nørrebro Odense Pia_Kjærsgaard SF skuespiller Thorkild_Simonsen tørklæde udlændingepolitik Vollsmose Århus Figure 27: Visualization map for Immigration in 1999 54 Visualizing News Data Streams Immigration 2000 andengenerationsindvandrere arj Berlingske_Tidende Bispehaven danskere Dansk_Folkeparti Esbjerg Folketinget halalhippier indvandrerbaggrund Instr Karen_Jespersen Khader Københavns_Universitet Nyrup Nørrebro Odense Pia_Kjærsgaard SF Socialdemokratiet tørklæde Venstre Århus Figure 28: Visualization map for Immigration in 2000 55 Visualizing News Data Streams Immigration 2007 danskere Dansk_Folkeparti DF Fogh Folketinget Gellerup Henrik indvandrerbaggrund Khader Marianne_Jelved Marwan Ny_Alliance Nørrebro PET romaer Sarkozy SF skuespiller tørklæde Venstre Århus Özlem Figure 35: Visualization map for Immigration in 2007 62 Visualizing News Data Streams Immigration 2008 Berlingske_Tidende Carlos danskere Dansk_Folkeparti DF drenge Fogh Folketinget Frederik Hells_Angels indvandrerbaggrund islam McCain muslimer Nørrebro Obama Pia_Kjærsgaard romaer Sarkozy SF tørklæde vampyr Villy_Søvndal Århus Figure 36: Visualization map for Immigration in 2008 63 Visualizing News Data Streams Immigration 2009 bandekrigen danskere Dansk_Folkeparti DF Hans_Lassen Hells_Angels ikke−vestlige indvandrerbaggrund Jønke Khader Mustafa Nørrebro Obama PET Pia_Kjærsgaard rockere romaer SF skyderier Suleebo Tingbjerg tørklæde uighurer Århus Figure 37: Visualization map for Immigration in 2009 64 Visualizing News Data Streams Immigration 2010 Arizona Chopin danskere Dansk_Folkeparti DF ikke−vestlige indvandrerbaggrund Josef Kenneth Københavns_Universitet Madsen Malmø museum mårhunde Nørrebro Obama Pia_Kjærsgaard romaer Rosengård sammenhængskraft SF Sverigedemokraterne udlændingepolitik Århus Figure 38: Visualization map for Immigration in 2010 65 Visualizing News Data Streams Immigration 2011 Aarhus blog Danmarks_Statistik danskere Dansk_Folkeparti DF drabsforsøg DSK Folketinget Hells_Angels ikke−vestlige indvandrerbaggrund Jamal kronik MFP Nørrebro ORG PET rockere SF Strauss−Kahn stuepige Søren_Pind tosprogede udlændingepolitik Figure 39: Visualization map for Immigration in 2011 66 Visualizing News Data Streams Immigration 2012 ● Aarhus danskere Dansk_Folkeparti DF dingoen Eske filmen flygel Hamid_Rahmati ikke−vestlige indvandrerbaggrund Kristian_Thulesen_Dahl Mir Nørrebro Obama Pia_Kjærsgaard Reich Romney Sarkozy skuespiller Sleiman Wilders Figure 40: Visualization map for Immigration in 2012 67 Visualizing News Data Streams Immigration 2013 Aarhus danskere Dansk_Folkeparti DF digte Fremskridtspartiet Gyldent_Daggry indvandrerbaggrund islam Khader Krasnik Københavns_Universitet Lars_Hedegaard muslimer Nyrup ph Republikanerne romaer SF tørklæde ulve Yahya_Hassan Figure 41: Visualization map for Immigration in 2013 68 Visualizing News Data Streams Immigration 2014 Aarhus Basim bogen danskere Dansk_Folkeparti DF digte elg Frank_Jensen Grønland ikke−vestlige indvandrerbaggrund kontanthjælp Krim Manu_Sareen Nørrebro Obama PET romaer SF skuespiller ulve velfærdsturisme Yahya_Hassan Figure 42: Visualization map for Immigration in 2014 69 Visualizing News Data Streams Immigration 2015 ● Charlie_Hebdo danskere Dansk_Folkeparti DF Dieudonné Egtvedpigen Facebook Folketinget Fremskridtspartiet Fryd Henrik indvandrerbaggrund Inger_Støjberg islam Islamisk_Stat Khader Kristian_Thulesen_Dahl kronik Lars_Løkke_Rasmussen Majid Manu_Sareen Nørrebro Pia_Kjærsgaard Venstre Figure 43: Visualization map for Immigration in 2015 70 View publication statsView publication stats