Multimedia Analytics
Newdle: Interactive Visual Exploration of Large Online News Collections Jing Yang, Dongning Luo, and Yujie Liu ■ University of North Carolina at Charlotte
N
ewspapers, television channels, and independent journalists release large numbers of news articles on the Internet. Online news provides timely, ambient information to not only the masses but also business and political policy makers, social scientists, and analysts in other application domains. Online news is also an overwhelming information source. For example, the New York Times Newdle is a visual-analytics (NYT; www.nytimes.com) delivsystem for exploring tagged ers thousands of news articles in online news collections. A less than one month. topic overview lets users grasp This extremely large volume of a large collection’s contents online news has created an urat a glance. Users can also gent need for tools that let users effectively and efficiently browse conduct in-depth analyses on topics, detect temporal trends, topics, tags, and articles. and search news of interest. Toward this end, researchers have developed many text analysis approaches, such as automatic news summarization,1 to extract valuable information from large online news collections. However, even after such analyses, the amount of deliverable information is still huge. Effectively and efficiently transferring large amounts of automatic-analysis results is challenging. Visual-analytics approaches, which tightly integrate automatic analysis and interactive visual exploration,2 offer a promising way to address this challenge. These approaches’ interactive visual interfaces exploit human perception abilities and domain knowledge together with computational powers to facilitate reasoning. Although a few on32
September/October 2010
line newsreaders employ automatic text analysis and use visualizations as their main interfaces, far more research is necessary to promote visual analytics in online-news exploration. To address the challenges faced by these approaches, we propose Newdle, a visual-analytics system for exploring tagged online news collections. “Newdle” stands for News Wordles, because wordles are its major visual metaphor.3 Wordles are attractive, cloud-like visualizations that pack many tags (word labels) with varying font sizes and colors into a small screen space. We’ve implemented a fully working prototype of Newdle, using the NYT RSS feeds as the example data source. Our case studies illustrate that Newdle is fun to use, effective, and efficient.
Challenges Addressed by Newdle Several challenges need to be addressed by onlinenews visual-analytics approaches. The first challenge is to use intuitive visual metaphors. Most persons who read or analyze online news aren’t visualization experts. The visual metaphors used for online-news visual-analytics approaches must be intuitive to novice users. The second challenge is to convey semantic information. Online news is unlike numeric data in that it contains rich semantic information. Effective online-news visual-analytics approaches must transfer a large amount of semantic information to users in an instant. The third challenge is to be scalable. Visualanalytics approaches should help users explore
Published by the IEEE Computer Society
0272-1716/10/$26.00 © 2010 IEEE
large online news collections through informationintensive displays and interactive exploration powered by automatic analyses. The final challenge is to support interactive, indepth analysis. Visual-analytics approaches should let users conduct a wide range of in-depth analyses according to their particular interests. Newdle addresses these challenges by using wordles to visually present semantic information of automatically generated news article clusters and provide a rich set of interactions that let users conduct in-depth visual analyses on topics, tags, and news articles. Wordles are easy to understand, fun to use, and attractive, making them intuitive visual metaphors. In addition, wordles display significant tags of article clusters and can quickly provide semantic information for them. Moreover, the integration of space-efficient wordles and clustering techniques lets Newdle scale to large amounts of information. Finally, interactions with wordles and other visualizations help users easily perform visual-analytics tasks.
date time so that it won’t retrieve data already in the database. Users can also define an article life threshold. The data extraction component will remove from the database all news entries older than this threshold.
How Newdle Works
the networks to detect hot topics and relations between tags and articles. It uses the results to construct visualizations in the visualization-andinteraction component.
Newdle is a C++ program that uses wxWidgets for its GUI, and MySQL for its data storage. Its three main components handle data extraction, network construction and analysis, and visualization and interaction. Here, we assume the news articles are accurately described by tags. This assumption is reasonable because many online news sources provide manually generated tags for their news articles in the RSS entries. We could easily extend our approach to visualize untagged news by integrating existing entity extraction and document summarization algorithms into it.
Data Extraction Newdle automatically fetches news article entries from NYT RSS feeds through Google Reader (www.reader.google.com). Google Reader provides a standardized XML format for RSS entries and caches them for up to 30 days. Newdle uses an XML parser to analyze the retrieved entries and get the tags, titles, time stamps, and summaries for articles and hyperlinks to the articles. This information is stored in a MySQL database and used by the other two components. Newdle can populate its database manually or automatically. In manual mode, Newdle updates the database by fetching the NYT entries upon a user’s request. In automatic mode, Newdle updates the database at user-defined time intervals. Newdle keeps track of the most recent up
Network Construction and Analysis Newdle constructs an article network and a bipartite article and tag network (article-tag network) to support interactive visual exploration of news articles. These networks explicitly define the relations among tags and articles. Newdle conducts clustering and automatic path analyses on
Online news provides timely, ambient information to not only the masses but also business and political policy makers, social scientists, and analysts.
Article network. This network describes the relations between articles. Each article is a vertex. An undirected, unweighted edge exists between two articles if and only if they share more tags than the article relation (AR) threshold. This edge indicates that the articles are directly related. Users can interactively change the AR threshold to define the relations more strictly or more loosely, which will lead to topics of finer or coarser granularities. Of course, we could measure the relations in other ways. We chose this simple approach because this measure is easy to understand and facilitates the changing of topic granularities. We conduct graph clustering on the article network. This involves grouping the graph’s vertices into clusters, taking into consideration the graph’s edge structure such that there are many edges in each cluster but relatively few edges between clusters.4 Our prototype uses the clustering algorithm provided by igraph (http://igraph.sourceforge.net), but any scalable, effective graph-clustering algorithm will do. Newdle considers a cluster to be a topic. A topic’s semantics constitute the most shared tags among the articles in the cluster. A topic’s temporal feature is the number of articles in this cluster over time. The relevance between topics A and B is the number IEEE Computer Graphics and Applications
33
Multimedia Analytics
Figure 1. The topic overview with coarse granularity (the article relation threshold is 2). The figure shows the 12 topics comprising the most articles. Users can access more topics via the scrolling bar on the display’s right.
of edges connecting the vertices in A to those in B. The distance between articles A and B is the length of the shortest path between them in the article network. Newdle calculates this distance upon requests from user interactions. Two articles are considered closely related if the distance between them is small. Article-tag network. This network describes the relations between tags and articles. It connects an article to each of its tags. A tag’s relevant articles are all those that connect to it. To find the tags relevant to a given tag, we select all the articles connected to it and then find all the tags connected to them.
Visualization and Interaction Using wordles as the basic visual metaphor, Newdle presents three basic views: the category index, the topic overview, and the detail view. These views let users perform several types of interactive visual analysis. Wordles. The wordles we use have strong semantic meanings. First, each wordle represents a news topic’s semantics—namely, the most shared tags. Those tags convey the topic’s “who, what, when, why, and where” so that users can learn this information at a glance. The individual tags’ semantic meanings are clear in such a context. Using wordles in this way significantly distinguishes our approach from the many approaches that use tag clouds to display a set of tags that aren’t necessarily related. 34
September/October 2010
Second, our wordles distinguish between different categories of tags. In NYT news articles, each tag has a category, such as location, person, organization, and topic (the NYT provides the categorization). We assign a unique color to all tags of the same category so that users can focus on a particular category’s tags during visual exploration. We use a qualitative color scheme from Colorbrewer (www.colorbrewer2.org). As Colorbrewer states, its qualitative schemes are best suited for representing nominal or categorical data; the hues create the primary visual difference between classes. For example, in Figure 1, all location tags are gold, and all person tags are purple. A user interested in location tags can thus look for gold tags. Third, a tag’s size in a wordle is proportional to the number of articles with this tag in the topic. Thus, a topic’s significant tags easily catch the user’s eyes. To reveal a topic’s temporal trends, we overlay a line graph on a wordle to indicate the daily number of articles published on this topic (see Figure 1). The line graphs in different wordles are in the same scale so that users can compare the topics with regard to their strengths over time. For example, we can immediately notice significant, bursty topics in Figure 1. The category index. In this view, users begin their visual exploration by setting the time period, the AR threshold, and the news category. Users can set the time period and the AR threshold through the category index’s menu bar. The default period is from 30 days before the present,
Figure 2. The topic overview with fine granularity (the article relation threshold is 3). We’ve conducted a relevant tag selection for the tag Haiti. The yellow background indicates the selected tags, which highlight topics related to Haiti.
during which the news entries are guaranteed to be accessible. Users can interactively change the time period. It can be within or beyond the default, provided the data is in the database. The default AR threshold is 2. In other words, Newdle considers articles sharing two or more tags to be directly related. As we mentioned earlier, users can change this threshold for different topic granularities. For example, Figure 2 shows the same news collection as in Figure 1 but with a finer granularity. The category index lists news categories from NYT RSS feeds, including world news, US news, business news, technology news, sports news, and health news. Newdle displays a wordle for each category to help users select a category of interest. It displays the most frequent tags in that category during the selected time period. When users click on a category name, Newdle defines a news collection and displays its topic overview. Users can always go back to the category index to change the settings and start a new exploration. Throughout the rest of this article, when we refer to a news collection, we mean the collection defined by the current time period and the category of interest. The topic overview. This view lets users browse significant topics to learn their semantics and temporal features at a glance. From this view, users can conduct or trigger many interactive explorations, such as topic or tag analysis. To generate this view, Newdle constructs an ar
ticle network using the given AR threshold. Clustering takes place on the article network. Newdle sorts the resulting topics in descending order according to the number of articles in them. It then generates wordles for the topics with the most articles and displays them in the overview, line by line, following the order. Figures 1 and 2 show topic overviews. The detail view. This view has multiple variations serving different purposes. Its basic form (see Figure 3) has multiple rows, each for a different topic. On a row’s left is the topic’s wordle; on the right is an HTML box listing the articles’ titles and tags. The titles’ colors range from white to blue to indicate their ages: the bluer a title is, the older the article is. Newdle lists the titles in ascending order according to their age. When users click the show detail tag in an HTML box, this box displays more information about the articles, such as their time stamps and summaries. Users can click on a title to open a Web browser window, which will display the original news article. Besides the basic form, the detail view presents results for topic investigations, article searches, tag comparisons, and extended-reading searches (see Figures 4 through 7). We describe these tasks more in the next section. Interactive visual analysis. Newdle helps users conduct in-depth visual analyses on topics, tags, and news articles. Organizing news articles into topics IEEE Computer Graphics and Applications
35
Multimedia Analytics
Figure 3. The detail view of significant topics. Users can examine the clusters’ semantics on the left and access individual news articles on the right.
Figure 4. A detail view showing results for topic investigation. The top row displays the topic of interest, and the other rows display topics closely related to it.
lets users explore a large news collection effectively and efficiently. To let users analyze topics, Newdle provides four interactions. The first is topic browsing in the topic overview. We fix the wordles’ sizes so that the tags in them will be readable. When there are more wordles than a screen can hold, a scrolling bar appears so that users can navigate an area larger than the screen space. The second interaction is topic browsing in the 36
September/October 2010
detail view. The detail view (see Figure 3) displays the significant topics in the same order as in the topic overview. Users can browse news articles’ titles in the topics, access news articles, and start extended-reading searches. The third interaction is topic reconstruction. If users are unsatisfied with the topics’ granularity, they can adjust the AR threshold. For example, if they want to see more details in the overview, they can increase the AR threshold to construct topics
Figure 5. A detail view showing article search results. The rows display articles with the tag Haiti, organized by their topics.
Figure 6. A detail view showing the comparison between two tags of interest. The top row displays the tags of interest. The other rows display articles with either tag or both of the tags, organized by their topics. The flags in front of each article indicate how the tags appear in the article.
with more coherent content. Figures 1 and 2 show topic overviews of the same news collection with different AR thresholds. The fourth interaction is topic investigation. Users can examine a topic in detail and investigate its relevant topics in the topic investigation view (see Figure 4). They can trigger this view by doubleclicking on a topic from the topic overview. In this
view, the topic of interest (TOI) is in the top row. The other rows display topics relevant to the TOI, sorted by their relevance. In these topics’ HTML boxes, stars appear before the titles, indicating how the articles relate to the TOI. Three stars mean the article is directly connected to one or more articles in the TOI. One or two stars indicate an indirect connection with IEEE Computer Graphics and Applications
37
Multimedia Analytics
Figure 7. A detail view showing extended-reading search results. The top row displays the article of interest. The other rows display articles directly or indirectly connected to it, organized by their topics.
Users can find significant tags by looking for large tags from the wordles in the topic overview. a distance of 3 or 2, respectively, in the article network. Newdle sorts the titles first by their stars, then by their ages. Users can find significant tags by looking for large tags from the wordles in the topic overview. Besides this manual approach, Newdle provides four interactions for conducting in-depth tag analyses. First, users can select a set of tags from the wordles and conduct operations on that selection. Selected tags are highlighted by red surrounding boxes. Users can manually select or unselect a tag. When the user’s cursor hovers over a tag, a white surrounding box appears. Users can then press the mouse button to change the tag’s selection status. Newdle can also automatically select tags relevant to a set of tags of interest. To trigger this operation, users add the tags of interest to the current selection and then click a relevant-tagsearch button. The system uses the article-tag network to find articles with the tags of interest. Then, for each article found, the system adds to the current selection its tags in the wordle of the topic to which it belongs. In Figure 2, we’ve selected the tags with the yellow background, using a relevant-tag search starting from the tag Haiti. 38
September/October 2010
The automatic selection can choose the same tag in multiple wordles—namely, in different semantic contexts. A tag’s selection in one wordle doesn’t mean that it’s selected in other wordles. Users can clear the selection by clicking a button. Second, users can click a button to search articles with all or any tags in the current selection. The results appear in the search result view. In this view, each row is a topic containing one or more resulting articles, sorted in descending order according to the number of search results they contain. The HTML boxes display only the resulting articles, with the searched tags italicized. Figure 5 displays search results for the tag Haiti. Third, for marking selected tags, Newdle provides several buttons with different colors. When users click such a button, the tags’ background color in the current selection changes to the button’s color. For example, in Figure 2, the tags related to Haiti have a yellow background. Users can begin a new search after they mark the tags in the current selection. Thus, they can compare the results of multiple selections by marking those results in different colors. Marked tags can also help users find TOIs. Fourth, when users have selected two tags, they can click a button to trigger a tag comparison in order to find the relation between the tags. Figure 6 shows this view for the tags Yemen and Abdulmutallab, Umar Farouk. On the top, the two compared tags are in different colors. Let’s call the tag on the left “tag 1,” and the other one “tag 2.” Newdle displays articles having either tag or both of the
tags in their topics. In the HTML boxes, two circles appear in front of each title. The left one is filled with tag 1’s color if the article includes tag 1, and is empty if the article doesn’t include tag 1. In the same way, the right one indicates whether the article includes tag 2. To conduct an extended-reading search, users first indicate an article of interest (AOI) by clicking the button on an article’s left in detail view. They then click another button to trigger the extended-reading list for that article (see Figure 7). The top row displays the AOI; the other rows display articles directly or indirectly connected to the AOI, organized by their topics. The stars indicate an article’s distance to the AOI. One, two, or three stars indicate a distance of 3, 2, or 1. In other words, an article with three stars is directly related to the AOI, and an article with one or two stars is indirectly related.
Case Studies Our two case studies explored the NYT world news from 17 December 2009 to 19 January 2010. This collection comprised 917 articles.
Haiti Earthquake This case study started with the topic overview in Figure 1. The top-right wordle caught our attention because there was a big burst in its line graph. By reading its tags, we knew it was about the Haiti earthquake. We wanted to refine the topic granularity, so we changed the AR threshold from 2 to 3. This interaction generated a new topic overview. We decided to search and examine topics about Haiti from this view. We began by highlighting the topics related to Haiti. To do this, we clicked the tag Haiti in a wordle, conducted a relevant-tag search, and marked the selected tags with the yellow background. Figure 2 shows the results. We noticed multiple significant topics highlighted by their tags’ yellow background. The first topic was about the disasters and emergencies in Haiti. The second was about humanitarian aid to Haiti. The third concerned the latest update about the earthquake. We then clicked these topics one by one to examine them in detail. We also performed this task in a different way. We began by selecting Haiti from the topic overview and searching articles with this tag. Figure 5 shows the results. We found all the topics we noticed earlier. We also browsed the topics’ article titles and clicked on those of interest to read the articles.
Christmas Day Bombing Attempt This case study investigated the topic that’s second
from the left in Figure 2’s top row, which seemed to be breaking news according to its line graph. We first clicked the topic’s wordle from the topic overview to conduct a topic investigation (see Figure 4). The TOI is in the top row. From the line graph in the wordle, we noticed that a news burst occurred some time after 17 December 2009. We dragged the scrolling bar of this topic’s HTML box to examine the earliest news article,
Our two case studies explored the New York Times world news from 17 December 2009 to 19 January 2010. This collection comprised 917 articles. which was published 26 December 2009. We clicked the title to access the article. It reported on a bombing attempt on a Christmas Day flight to Detroit that was prevented by the passengers and crew members. We then examined more articles in this topic. Most were follow-ups of this event, along with articles discussing security and warning systems, airports, and airlines, as suggested by the wordle. We then browsed the relevant topics under Figure 4’s top row. The topic in the second row discussed Yemen, terrorism, and Al Qaeda. The topic in the third row was about how President Obama and the US reacted to the bombing attempt. We were curious about why Yemen was related to the bombing attempt. To investigate this relationship, we selected the tags Yemen and Abdulmutallab, Umar Farouk and compared them. The latter is a person’s name (which we know from the tag’s color) that appears in both the TOI and the topic in the second row of Figure 4. In the tag comparison results in Figure 6, several articles with both tags popped up. By reading the news with the title “Yemen Says Bomb Suspect Met with Qaeda Figures,” we learned that Umar Farouk Abdulmutallab made the bombing attempt and that he met with Al Qaeda operatives in Yemen before his journey. To investigate the event in a different way, we started from the basic detail view (see Figure 3). We selected the earliest news article in this topic and conducted an extended-reading search for this article. Figure 7 shows the results. Interestingly, according to the line graphs, the topic in the third row had appeared before the bombing attempt. By examining the HTML box, we noticed that some IEEE Computer Graphics and Applications
39
Multimedia Analytics
Related Work on Text Visual Analytics
N
ewdle is related to two categories of previous research: online-news visual exploration and visual analytics of document collections.
Online-News Visual Exploration Newsmap (http://newsmap.jp) uses a tree map to illustrate automatic news article clustering results from Google News (http://news.google.com). Each cell in the tree map represents a news cluster; a cell’s size indicates the number of related articles in the cluster. Article titles appear in the cells. Google News Timeline (http://newstimeline.googlelabs. com) organizes search results chronologically. This Web application lets users view news and other data sources in a browsable, graphical timeline. The Narratives application tracks references to a news article in social media (such as blogs).1 It uses line graphs to visualize how concepts have changed over time and how they relate to one another. VisGets provides coordinated geospatial, temporal, and topical views to let users interactively explore online news.2 Users can find topics of interest through dynamicsearch queries. FeedDemon (www.feeddemon.com) is commercial software that lets users read, tag, watch for, search, and share news. Newdle (see the main article) differs from these approaches in that it provides a rich set of interactions to support in-depth analysis, using an underlying article network and article-tag network.
Visual Analytics of Document Collections In-Spire employs clustering analysis and uses mountain and galaxy metaphors to depict hot topics.3 It displays those topics’ semantic information as labels. ThemeRiver uses a river metaphor to depict significant keywords’ thematic strengths.4 Trist (The Rapid Information Scanning Tool) lets analysts formulate, refine, organize, and execute queries over large document collections.5 Its multipane view lets users examine search results from different perspectives, such as clustering, trend analysis, comparisons, and differences.
Jigsaw provides multiple coordinated views of document entities, illustrating connections between entities across different documents.6 Coplink presents a hyperbolic-tree view and a spring-embedded graph layout of relevant entities to help law enforcement officials extract information from police case reports and analyze criminal networks.7 Unlike these approaches, Newdle presents the major topics in a document collection and their temporal trends using wordles combined with temporal overlay lines. (For more on wordles, see the main article.) It supports a range of visual-analytics tasks, exploiting the underlying article and article-tag networks without explicitly displaying them.
References 1. D. Fisher et al., “Narratives: A Visualization to Track Narrative Events as They Develop,” Proc. IEEE Symp. Visual Analytics Science and Technology (VAST 08), IEEE CS Press, 2008, pp. 115–122. 2. M. Dörk et al., “VisGets: Coordinated Visualizations for Web-Based Information Exploration and Discovery,” IEEE Trans. Visualization and Computer Graphics, vol. 14, no. 6, 2008, pp. 1205–1212. 3. J.A. Wise et al., “Visualizing the Non-visual: Spatial Analysis and Interaction with Information from Text Documents,” Proc. IEEE Symp. Information Visualization, IEEE CS Press, 1995, pp. 51–58. 4. S. Havre et al., “ThemeRiver: Visualizing Thematic Changes in Large Document Collections,” IEEE Trans. Visualization and Computer Graphics, vol. 8, no. 1, 2002, pp. 9–20. 5. D. Jonker et al., “Information Triage with Trist,” Proc. Int’l Conf. Intelligence Analysis, Oculus Info, 2005; www.oculusinfo. com/papers/Oculus_TRIST_Final_Distrib.pdf. 6. J. Stasko, C. Görg, and Z. Liu, “Jigsaw: Supporting Investiga tive Analysis through Interactive Visualization,” Information Visualization, vol. 7, no. 2, 2008, pp. 118–132. 7. H. Chen et al., “Coplink Connect: Information and Knowl edge Management for Law Enforcement,” Decision Support Systems, vol. 34, no. 3, 2003, pp. 271–285.
articles indirectly related to the bombing attempt were published before 26 December 2009. Those articles discussed Yemen and Al Qaeda.
For information on previous research related to Newdle, see the “Related Work on Visual Analytics” sidebar.
N
Acknowledgments
ewdle isn’t limited to exploring online news collections; it also supports exploration of other document collections. We plan to use Newdle in several other applications, such as science policy analysis and email visualizations. We also plan to conduct formal user studies to evaluate its effectiveness and efficiency. 40
September/October 2010
This article is based on research supported by the US National Science Foundation under grant IIS0946400 and the US Department of Homeland Security under grant 2008-ST-108-000002. The views and conclusions are the authors’ and don’t necessarily represent the official policies, either expressed or im-
plied, of the US National Science Foundation or the US Department of Homeland Security.
References 1. K. Jones, “Automatic Summarizing: The State of the Art,” Information Processing and Management, vol. 43, no. 6, 2007, pp. 1449–1481. 2. J. Thomas and K. Cook, Illuminating the Path: The Research and Development Agenda for Visual Analytics, National Visualization and Analytics Center, US Dept. of Homeland Security, 2005. 3. F. Viégas, M. Wattenberg, and J. Feinberg, “Par ticipatory Visualization with Wordle,” IEEE Trans. Visualization and Computer Graphics, vol. 15, no. 6, 2009, pp. 1137–1144. 4. S.E. Schaeffer, “Graph Clustering,” Computer Science Rev., vol. 1, no. 1, 2007, pp. 27–64. Jing Yang is an assistant professor in the Computer Science Department at the University of North Carolina at Charlotte. She’s also a co-principal investigator for the US Department of Homeland Security Center of Excellence on Visual Analytics for Command, Control, and Interoperabil-
ity and the Southeastern Regional Visualization and Analytics Center. Her research interests include visual analytics and information visualization. Yang has a PhD in computer science from Worcester Polytechnic Institute. Contact her at [email protected]. Dongning Luo is pursuing a PhD in computer science at the University of North Carolina at Charlotte. His research interests include data and visual analytics on largescale text collections. Luo has a BE in electrical engineering from Shanghai Jiao Tong University. He’s a student member of the IEEE Computer Society. Contact him at dluo2@ uncc.edu. Yujie Liu is pursuing a PhD in computer science at the University of North Carolina at Charlotte. Her research interests include visual analytics on large-scale text collections and evaluations. Liu has an MS in automatic control from Huazhong University of Science and Technology. She’s a student member of the IEEE Computer Society. Contact her at [email protected]. Selected CS articles and columns are also available for free at http://ComputingNow.computer.org.
EXECUTIVE STAFF PURPOSE: The IEEE Computer Society is the world’s largest association of computing professionals and is the leading provider of technical information in the field. MEMBERSHIP: Members receive the monthly magazine Computer, discounts, and opportunities to serve (all activities are led by volunteer members). Membership is open to all IEEE members, affiliate society members, and others interested in the computer field. COMPUTER SOCIETY WEB SITE: www.computer.org OMBUDSMAN: Email [email protected]. Next Board Meeting: 15–16 Nov. 2010, New Brunswick, NJ, USA
EXECUTIVE COMMITTEE President: James D. Isaak* President-Elect: Sorel Reisman;* Past President: Susan K. (Kathy) Land, CSDP;* VP, Standards Activities: Roger U. Fujii (1st VP);* Secretary: Jeffrey M. Voas (2nd VP);* VP, Educational Activities: Elizabeth L. Burd;* VP, Member & Geographic Activities: Sattupathu V. Sankaran;† VP, Publications: David Alan Grier;* VP, Professional Activities: James W. Moore;* VP, Technical & Conference Activities: John W. Walz;* Treasurer: Frank E. Ferrante;* 2010–2011 IEEE Division V Director: Michael R. Williams;† 2009–2010 IEEE Division VIII Director: Stephen L. Diamond;† 2010 IEEE Division VIII Director-Elect: Susan K. (Kathy) Land, CSDP;* Computer Editor in Chief: Carl K. Chang† *voting member of the Board of Governors
†nonvoting member of the Board of Governors
BOARD OF GOVERNORS Term Expiring 2010: Piere Bourque; André Ivanov; Phillip A. Laplante; Itaru Mimura; Jon G. Rokne; Christina M. Schober; Ann E.K. Sobel Term Expiring 2011: Elisa Bertino, George V. Cybenko, Ann DeMarle, David S. Ebert, David A. Grier, Hironori Kasahara, Steven L. Tanimoto Term Expiring 2012: Elizabeth L. Burd, Thomas M. Conte, Frank E. Ferrante, Jean-Luc Gaudiot, Luis Kun, James W. Moore, John W. Walz
Executive Director: Angela R. Burgess; Associate Executive Director; Director, Governance: Anne Marie Kelly; Director, Finance & Accounting: John Miller; Director, Membership Development: Violet S. Doan; Director, Products & Services: Evan Butterfield; Director, Sales & Marketing: Dick Price
COMPUTER SOCIETY OFFICES Washington, D.C.: 2001 L St., Ste. 700, Washington, D.C. 20036 Phone: +1 202 371 0101; Fax: +1 202 728 9614; Email: [email protected] Los Alamitos: 10662 Los Vaqueros Circle, Los Alamitos, CA 90720-1314 Phone: +1 714 821 8380; Email: [email protected] Membership & Publication Orders: Phone: +1 800 272 6657; Fax: +1 714 821 4641; Email: [email protected] Asia/Pacific: Watanabe Building, 1-4-2 Minami-Aoyama, Minato-ku, Tokyo 107-0062, Japan Phone: +81 3 3408 3118 • Fax: +81 3 3408 3553 Email: [email protected]
IEEE OFFICERS President: Pedro A. Ray; President-Elect: Moshe Kam; Past President: John R. Vig; Secretary: David G. Green; Treasurer: Peter W. Staecker; President, Standards Association Board of Governors: ; W. Charlston Adams; VP, Educational Activities: Tariq S. Durrani; VP, Membership & Geographic Activities: Barry L. Shoop; VP, Publication Services & Products: Jon G. Rokne; VP, Technical Activities: Roger D. Pollard; IEEE Division V Director: Michael R. Williams; IEEE Division VIII Director: Stephen L. Diamond; President, IEEE-USA: Evelyn H. Hirt
revised 17 Jun. 2010
IEEE Computer Graphics and Applications
41
Newdle: Interactive Visual Exploration of Large Online News Collections Jing Yang, Dongning Luo, and Yujie Liu ■ University of North Carolina at Charlotte
N
ewspapers, television channels, and independent journalists release large numbers of news articles on the Internet. Online news provides timely, ambient information to not only the masses but also business and political policy makers, social scientists, and analysts in other application domains. Online news is also an overwhelming information source. For example, the New York Times Newdle is a visual-analytics (NYT; www.nytimes.com) delivsystem for exploring tagged ers thousands of news articles in online news collections. A less than one month. topic overview lets users grasp This extremely large volume of a large collection’s contents online news has created an urat a glance. Users can also gent need for tools that let users effectively and efficiently browse conduct in-depth analyses on topics, detect temporal trends, topics, tags, and articles. and search news of interest. Toward this end, researchers have developed many text analysis approaches, such as automatic news summarization,1 to extract valuable information from large online news collections. However, even after such analyses, the amount of deliverable information is still huge. Effectively and efficiently transferring large amounts of automatic-analysis results is challenging. Visual-analytics approaches, which tightly integrate automatic analysis and interactive visual exploration,2 offer a promising way to address this challenge. These approaches’ interactive visual interfaces exploit human perception abilities and domain knowledge together with computational powers to facilitate reasoning. Although a few on32
September/October 2010
line newsreaders employ automatic text analysis and use visualizations as their main interfaces, far more research is necessary to promote visual analytics in online-news exploration. To address the challenges faced by these approaches, we propose Newdle, a visual-analytics system for exploring tagged online news collections. “Newdle” stands for News Wordles, because wordles are its major visual metaphor.3 Wordles are attractive, cloud-like visualizations that pack many tags (word labels) with varying font sizes and colors into a small screen space. We’ve implemented a fully working prototype of Newdle, using the NYT RSS feeds as the example data source. Our case studies illustrate that Newdle is fun to use, effective, and efficient.
Challenges Addressed by Newdle Several challenges need to be addressed by onlinenews visual-analytics approaches. The first challenge is to use intuitive visual metaphors. Most persons who read or analyze online news aren’t visualization experts. The visual metaphors used for online-news visual-analytics approaches must be intuitive to novice users. The second challenge is to convey semantic information. Online news is unlike numeric data in that it contains rich semantic information. Effective online-news visual-analytics approaches must transfer a large amount of semantic information to users in an instant. The third challenge is to be scalable. Visualanalytics approaches should help users explore
Published by the IEEE Computer Society
0272-1716/10/$26.00 © 2010 IEEE
large online news collections through informationintensive displays and interactive exploration powered by automatic analyses. The final challenge is to support interactive, indepth analysis. Visual-analytics approaches should let users conduct a wide range of in-depth analyses according to their particular interests. Newdle addresses these challenges by using wordles to visually present semantic information of automatically generated news article clusters and provide a rich set of interactions that let users conduct in-depth visual analyses on topics, tags, and news articles. Wordles are easy to understand, fun to use, and attractive, making them intuitive visual metaphors. In addition, wordles display significant tags of article clusters and can quickly provide semantic information for them. Moreover, the integration of space-efficient wordles and clustering techniques lets Newdle scale to large amounts of information. Finally, interactions with wordles and other visualizations help users easily perform visual-analytics tasks.
date time so that it won’t retrieve data already in the database. Users can also define an article life threshold. The data extraction component will remove from the database all news entries older than this threshold.
How Newdle Works
the networks to detect hot topics and relations between tags and articles. It uses the results to construct visualizations in the visualization-andinteraction component.
Newdle is a C++ program that uses wxWidgets for its GUI, and MySQL for its data storage. Its three main components handle data extraction, network construction and analysis, and visualization and interaction. Here, we assume the news articles are accurately described by tags. This assumption is reasonable because many online news sources provide manually generated tags for their news articles in the RSS entries. We could easily extend our approach to visualize untagged news by integrating existing entity extraction and document summarization algorithms into it.
Data Extraction Newdle automatically fetches news article entries from NYT RSS feeds through Google Reader (www.reader.google.com). Google Reader provides a standardized XML format for RSS entries and caches them for up to 30 days. Newdle uses an XML parser to analyze the retrieved entries and get the tags, titles, time stamps, and summaries for articles and hyperlinks to the articles. This information is stored in a MySQL database and used by the other two components. Newdle can populate its database manually or automatically. In manual mode, Newdle updates the database by fetching the NYT entries upon a user’s request. In automatic mode, Newdle updates the database at user-defined time intervals. Newdle keeps track of the most recent up
Network Construction and Analysis Newdle constructs an article network and a bipartite article and tag network (article-tag network) to support interactive visual exploration of news articles. These networks explicitly define the relations among tags and articles. Newdle conducts clustering and automatic path analyses on
Online news provides timely, ambient information to not only the masses but also business and political policy makers, social scientists, and analysts.
Article network. This network describes the relations between articles. Each article is a vertex. An undirected, unweighted edge exists between two articles if and only if they share more tags than the article relation (AR) threshold. This edge indicates that the articles are directly related. Users can interactively change the AR threshold to define the relations more strictly or more loosely, which will lead to topics of finer or coarser granularities. Of course, we could measure the relations in other ways. We chose this simple approach because this measure is easy to understand and facilitates the changing of topic granularities. We conduct graph clustering on the article network. This involves grouping the graph’s vertices into clusters, taking into consideration the graph’s edge structure such that there are many edges in each cluster but relatively few edges between clusters.4 Our prototype uses the clustering algorithm provided by igraph (http://igraph.sourceforge.net), but any scalable, effective graph-clustering algorithm will do. Newdle considers a cluster to be a topic. A topic’s semantics constitute the most shared tags among the articles in the cluster. A topic’s temporal feature is the number of articles in this cluster over time. The relevance between topics A and B is the number IEEE Computer Graphics and Applications
33
Multimedia Analytics
Figure 1. The topic overview with coarse granularity (the article relation threshold is 2). The figure shows the 12 topics comprising the most articles. Users can access more topics via the scrolling bar on the display’s right.
of edges connecting the vertices in A to those in B. The distance between articles A and B is the length of the shortest path between them in the article network. Newdle calculates this distance upon requests from user interactions. Two articles are considered closely related if the distance between them is small. Article-tag network. This network describes the relations between tags and articles. It connects an article to each of its tags. A tag’s relevant articles are all those that connect to it. To find the tags relevant to a given tag, we select all the articles connected to it and then find all the tags connected to them.
Visualization and Interaction Using wordles as the basic visual metaphor, Newdle presents three basic views: the category index, the topic overview, and the detail view. These views let users perform several types of interactive visual analysis. Wordles. The wordles we use have strong semantic meanings. First, each wordle represents a news topic’s semantics—namely, the most shared tags. Those tags convey the topic’s “who, what, when, why, and where” so that users can learn this information at a glance. The individual tags’ semantic meanings are clear in such a context. Using wordles in this way significantly distinguishes our approach from the many approaches that use tag clouds to display a set of tags that aren’t necessarily related. 34
September/October 2010
Second, our wordles distinguish between different categories of tags. In NYT news articles, each tag has a category, such as location, person, organization, and topic (the NYT provides the categorization). We assign a unique color to all tags of the same category so that users can focus on a particular category’s tags during visual exploration. We use a qualitative color scheme from Colorbrewer (www.colorbrewer2.org). As Colorbrewer states, its qualitative schemes are best suited for representing nominal or categorical data; the hues create the primary visual difference between classes. For example, in Figure 1, all location tags are gold, and all person tags are purple. A user interested in location tags can thus look for gold tags. Third, a tag’s size in a wordle is proportional to the number of articles with this tag in the topic. Thus, a topic’s significant tags easily catch the user’s eyes. To reveal a topic’s temporal trends, we overlay a line graph on a wordle to indicate the daily number of articles published on this topic (see Figure 1). The line graphs in different wordles are in the same scale so that users can compare the topics with regard to their strengths over time. For example, we can immediately notice significant, bursty topics in Figure 1. The category index. In this view, users begin their visual exploration by setting the time period, the AR threshold, and the news category. Users can set the time period and the AR threshold through the category index’s menu bar. The default period is from 30 days before the present,
Figure 2. The topic overview with fine granularity (the article relation threshold is 3). We’ve conducted a relevant tag selection for the tag Haiti. The yellow background indicates the selected tags, which highlight topics related to Haiti.
during which the news entries are guaranteed to be accessible. Users can interactively change the time period. It can be within or beyond the default, provided the data is in the database. The default AR threshold is 2. In other words, Newdle considers articles sharing two or more tags to be directly related. As we mentioned earlier, users can change this threshold for different topic granularities. For example, Figure 2 shows the same news collection as in Figure 1 but with a finer granularity. The category index lists news categories from NYT RSS feeds, including world news, US news, business news, technology news, sports news, and health news. Newdle displays a wordle for each category to help users select a category of interest. It displays the most frequent tags in that category during the selected time period. When users click on a category name, Newdle defines a news collection and displays its topic overview. Users can always go back to the category index to change the settings and start a new exploration. Throughout the rest of this article, when we refer to a news collection, we mean the collection defined by the current time period and the category of interest. The topic overview. This view lets users browse significant topics to learn their semantics and temporal features at a glance. From this view, users can conduct or trigger many interactive explorations, such as topic or tag analysis. To generate this view, Newdle constructs an ar
ticle network using the given AR threshold. Clustering takes place on the article network. Newdle sorts the resulting topics in descending order according to the number of articles in them. It then generates wordles for the topics with the most articles and displays them in the overview, line by line, following the order. Figures 1 and 2 show topic overviews. The detail view. This view has multiple variations serving different purposes. Its basic form (see Figure 3) has multiple rows, each for a different topic. On a row’s left is the topic’s wordle; on the right is an HTML box listing the articles’ titles and tags. The titles’ colors range from white to blue to indicate their ages: the bluer a title is, the older the article is. Newdle lists the titles in ascending order according to their age. When users click the show detail tag in an HTML box, this box displays more information about the articles, such as their time stamps and summaries. Users can click on a title to open a Web browser window, which will display the original news article. Besides the basic form, the detail view presents results for topic investigations, article searches, tag comparisons, and extended-reading searches (see Figures 4 through 7). We describe these tasks more in the next section. Interactive visual analysis. Newdle helps users conduct in-depth visual analyses on topics, tags, and news articles. Organizing news articles into topics IEEE Computer Graphics and Applications
35
Multimedia Analytics
Figure 3. The detail view of significant topics. Users can examine the clusters’ semantics on the left and access individual news articles on the right.
Figure 4. A detail view showing results for topic investigation. The top row displays the topic of interest, and the other rows display topics closely related to it.
lets users explore a large news collection effectively and efficiently. To let users analyze topics, Newdle provides four interactions. The first is topic browsing in the topic overview. We fix the wordles’ sizes so that the tags in them will be readable. When there are more wordles than a screen can hold, a scrolling bar appears so that users can navigate an area larger than the screen space. The second interaction is topic browsing in the 36
September/October 2010
detail view. The detail view (see Figure 3) displays the significant topics in the same order as in the topic overview. Users can browse news articles’ titles in the topics, access news articles, and start extended-reading searches. The third interaction is topic reconstruction. If users are unsatisfied with the topics’ granularity, they can adjust the AR threshold. For example, if they want to see more details in the overview, they can increase the AR threshold to construct topics
Figure 5. A detail view showing article search results. The rows display articles with the tag Haiti, organized by their topics.
Figure 6. A detail view showing the comparison between two tags of interest. The top row displays the tags of interest. The other rows display articles with either tag or both of the tags, organized by their topics. The flags in front of each article indicate how the tags appear in the article.
with more coherent content. Figures 1 and 2 show topic overviews of the same news collection with different AR thresholds. The fourth interaction is topic investigation. Users can examine a topic in detail and investigate its relevant topics in the topic investigation view (see Figure 4). They can trigger this view by doubleclicking on a topic from the topic overview. In this
view, the topic of interest (TOI) is in the top row. The other rows display topics relevant to the TOI, sorted by their relevance. In these topics’ HTML boxes, stars appear before the titles, indicating how the articles relate to the TOI. Three stars mean the article is directly connected to one or more articles in the TOI. One or two stars indicate an indirect connection with IEEE Computer Graphics and Applications
37
Multimedia Analytics
Figure 7. A detail view showing extended-reading search results. The top row displays the article of interest. The other rows display articles directly or indirectly connected to it, organized by their topics.
Users can find significant tags by looking for large tags from the wordles in the topic overview. a distance of 3 or 2, respectively, in the article network. Newdle sorts the titles first by their stars, then by their ages. Users can find significant tags by looking for large tags from the wordles in the topic overview. Besides this manual approach, Newdle provides four interactions for conducting in-depth tag analyses. First, users can select a set of tags from the wordles and conduct operations on that selection. Selected tags are highlighted by red surrounding boxes. Users can manually select or unselect a tag. When the user’s cursor hovers over a tag, a white surrounding box appears. Users can then press the mouse button to change the tag’s selection status. Newdle can also automatically select tags relevant to a set of tags of interest. To trigger this operation, users add the tags of interest to the current selection and then click a relevant-tagsearch button. The system uses the article-tag network to find articles with the tags of interest. Then, for each article found, the system adds to the current selection its tags in the wordle of the topic to which it belongs. In Figure 2, we’ve selected the tags with the yellow background, using a relevant-tag search starting from the tag Haiti. 38
September/October 2010
The automatic selection can choose the same tag in multiple wordles—namely, in different semantic contexts. A tag’s selection in one wordle doesn’t mean that it’s selected in other wordles. Users can clear the selection by clicking a button. Second, users can click a button to search articles with all or any tags in the current selection. The results appear in the search result view. In this view, each row is a topic containing one or more resulting articles, sorted in descending order according to the number of search results they contain. The HTML boxes display only the resulting articles, with the searched tags italicized. Figure 5 displays search results for the tag Haiti. Third, for marking selected tags, Newdle provides several buttons with different colors. When users click such a button, the tags’ background color in the current selection changes to the button’s color. For example, in Figure 2, the tags related to Haiti have a yellow background. Users can begin a new search after they mark the tags in the current selection. Thus, they can compare the results of multiple selections by marking those results in different colors. Marked tags can also help users find TOIs. Fourth, when users have selected two tags, they can click a button to trigger a tag comparison in order to find the relation between the tags. Figure 6 shows this view for the tags Yemen and Abdulmutallab, Umar Farouk. On the top, the two compared tags are in different colors. Let’s call the tag on the left “tag 1,” and the other one “tag 2.” Newdle displays articles having either tag or both of the
tags in their topics. In the HTML boxes, two circles appear in front of each title. The left one is filled with tag 1’s color if the article includes tag 1, and is empty if the article doesn’t include tag 1. In the same way, the right one indicates whether the article includes tag 2. To conduct an extended-reading search, users first indicate an article of interest (AOI) by clicking the button on an article’s left in detail view. They then click another button to trigger the extended-reading list for that article (see Figure 7). The top row displays the AOI; the other rows display articles directly or indirectly connected to the AOI, organized by their topics. The stars indicate an article’s distance to the AOI. One, two, or three stars indicate a distance of 3, 2, or 1. In other words, an article with three stars is directly related to the AOI, and an article with one or two stars is indirectly related.
Case Studies Our two case studies explored the NYT world news from 17 December 2009 to 19 January 2010. This collection comprised 917 articles.
Haiti Earthquake This case study started with the topic overview in Figure 1. The top-right wordle caught our attention because there was a big burst in its line graph. By reading its tags, we knew it was about the Haiti earthquake. We wanted to refine the topic granularity, so we changed the AR threshold from 2 to 3. This interaction generated a new topic overview. We decided to search and examine topics about Haiti from this view. We began by highlighting the topics related to Haiti. To do this, we clicked the tag Haiti in a wordle, conducted a relevant-tag search, and marked the selected tags with the yellow background. Figure 2 shows the results. We noticed multiple significant topics highlighted by their tags’ yellow background. The first topic was about the disasters and emergencies in Haiti. The second was about humanitarian aid to Haiti. The third concerned the latest update about the earthquake. We then clicked these topics one by one to examine them in detail. We also performed this task in a different way. We began by selecting Haiti from the topic overview and searching articles with this tag. Figure 5 shows the results. We found all the topics we noticed earlier. We also browsed the topics’ article titles and clicked on those of interest to read the articles.
Christmas Day Bombing Attempt This case study investigated the topic that’s second
from the left in Figure 2’s top row, which seemed to be breaking news according to its line graph. We first clicked the topic’s wordle from the topic overview to conduct a topic investigation (see Figure 4). The TOI is in the top row. From the line graph in the wordle, we noticed that a news burst occurred some time after 17 December 2009. We dragged the scrolling bar of this topic’s HTML box to examine the earliest news article,
Our two case studies explored the New York Times world news from 17 December 2009 to 19 January 2010. This collection comprised 917 articles. which was published 26 December 2009. We clicked the title to access the article. It reported on a bombing attempt on a Christmas Day flight to Detroit that was prevented by the passengers and crew members. We then examined more articles in this topic. Most were follow-ups of this event, along with articles discussing security and warning systems, airports, and airlines, as suggested by the wordle. We then browsed the relevant topics under Figure 4’s top row. The topic in the second row discussed Yemen, terrorism, and Al Qaeda. The topic in the third row was about how President Obama and the US reacted to the bombing attempt. We were curious about why Yemen was related to the bombing attempt. To investigate this relationship, we selected the tags Yemen and Abdulmutallab, Umar Farouk and compared them. The latter is a person’s name (which we know from the tag’s color) that appears in both the TOI and the topic in the second row of Figure 4. In the tag comparison results in Figure 6, several articles with both tags popped up. By reading the news with the title “Yemen Says Bomb Suspect Met with Qaeda Figures,” we learned that Umar Farouk Abdulmutallab made the bombing attempt and that he met with Al Qaeda operatives in Yemen before his journey. To investigate the event in a different way, we started from the basic detail view (see Figure 3). We selected the earliest news article in this topic and conducted an extended-reading search for this article. Figure 7 shows the results. Interestingly, according to the line graphs, the topic in the third row had appeared before the bombing attempt. By examining the HTML box, we noticed that some IEEE Computer Graphics and Applications
39
Multimedia Analytics
Related Work on Text Visual Analytics
N
ewdle is related to two categories of previous research: online-news visual exploration and visual analytics of document collections.
Online-News Visual Exploration Newsmap (http://newsmap.jp) uses a tree map to illustrate automatic news article clustering results from Google News (http://news.google.com). Each cell in the tree map represents a news cluster; a cell’s size indicates the number of related articles in the cluster. Article titles appear in the cells. Google News Timeline (http://newstimeline.googlelabs. com) organizes search results chronologically. This Web application lets users view news and other data sources in a browsable, graphical timeline. The Narratives application tracks references to a news article in social media (such as blogs).1 It uses line graphs to visualize how concepts have changed over time and how they relate to one another. VisGets provides coordinated geospatial, temporal, and topical views to let users interactively explore online news.2 Users can find topics of interest through dynamicsearch queries. FeedDemon (www.feeddemon.com) is commercial software that lets users read, tag, watch for, search, and share news. Newdle (see the main article) differs from these approaches in that it provides a rich set of interactions to support in-depth analysis, using an underlying article network and article-tag network.
Visual Analytics of Document Collections In-Spire employs clustering analysis and uses mountain and galaxy metaphors to depict hot topics.3 It displays those topics’ semantic information as labels. ThemeRiver uses a river metaphor to depict significant keywords’ thematic strengths.4 Trist (The Rapid Information Scanning Tool) lets analysts formulate, refine, organize, and execute queries over large document collections.5 Its multipane view lets users examine search results from different perspectives, such as clustering, trend analysis, comparisons, and differences.
Jigsaw provides multiple coordinated views of document entities, illustrating connections between entities across different documents.6 Coplink presents a hyperbolic-tree view and a spring-embedded graph layout of relevant entities to help law enforcement officials extract information from police case reports and analyze criminal networks.7 Unlike these approaches, Newdle presents the major topics in a document collection and their temporal trends using wordles combined with temporal overlay lines. (For more on wordles, see the main article.) It supports a range of visual-analytics tasks, exploiting the underlying article and article-tag networks without explicitly displaying them.
References 1. D. Fisher et al., “Narratives: A Visualization to Track Narrative Events as They Develop,” Proc. IEEE Symp. Visual Analytics Science and Technology (VAST 08), IEEE CS Press, 2008, pp. 115–122. 2. M. Dörk et al., “VisGets: Coordinated Visualizations for Web-Based Information Exploration and Discovery,” IEEE Trans. Visualization and Computer Graphics, vol. 14, no. 6, 2008, pp. 1205–1212. 3. J.A. Wise et al., “Visualizing the Non-visual: Spatial Analysis and Interaction with Information from Text Documents,” Proc. IEEE Symp. Information Visualization, IEEE CS Press, 1995, pp. 51–58. 4. S. Havre et al., “ThemeRiver: Visualizing Thematic Changes in Large Document Collections,” IEEE Trans. Visualization and Computer Graphics, vol. 8, no. 1, 2002, pp. 9–20. 5. D. Jonker et al., “Information Triage with Trist,” Proc. Int’l Conf. Intelligence Analysis, Oculus Info, 2005; www.oculusinfo. com/papers/Oculus_TRIST_Final_Distrib.pdf. 6. J. Stasko, C. Görg, and Z. Liu, “Jigsaw: Supporting Investiga tive Analysis through Interactive Visualization,” Information Visualization, vol. 7, no. 2, 2008, pp. 118–132. 7. H. Chen et al., “Coplink Connect: Information and Knowl edge Management for Law Enforcement,” Decision Support Systems, vol. 34, no. 3, 2003, pp. 271–285.
articles indirectly related to the bombing attempt were published before 26 December 2009. Those articles discussed Yemen and Al Qaeda.
For information on previous research related to Newdle, see the “Related Work on Visual Analytics” sidebar.
N
Acknowledgments
ewdle isn’t limited to exploring online news collections; it also supports exploration of other document collections. We plan to use Newdle in several other applications, such as science policy analysis and email visualizations. We also plan to conduct formal user studies to evaluate its effectiveness and efficiency. 40
September/October 2010
This article is based on research supported by the US National Science Foundation under grant IIS0946400 and the US Department of Homeland Security under grant 2008-ST-108-000002. The views and conclusions are the authors’ and don’t necessarily represent the official policies, either expressed or im-
plied, of the US National Science Foundation or the US Department of Homeland Security.
References 1. K. Jones, “Automatic Summarizing: The State of the Art,” Information Processing and Management, vol. 43, no. 6, 2007, pp. 1449–1481. 2. J. Thomas and K. Cook, Illuminating the Path: The Research and Development Agenda for Visual Analytics, National Visualization and Analytics Center, US Dept. of Homeland Security, 2005. 3. F. Viégas, M. Wattenberg, and J. Feinberg, “Par ticipatory Visualization with Wordle,” IEEE Trans. Visualization and Computer Graphics, vol. 15, no. 6, 2009, pp. 1137–1144. 4. S.E. Schaeffer, “Graph Clustering,” Computer Science Rev., vol. 1, no. 1, 2007, pp. 27–64. Jing Yang is an assistant professor in the Computer Science Department at the University of North Carolina at Charlotte. She’s also a co-principal investigator for the US Department of Homeland Security Center of Excellence on Visual Analytics for Command, Control, and Interoperabil-
ity and the Southeastern Regional Visualization and Analytics Center. Her research interests include visual analytics and information visualization. Yang has a PhD in computer science from Worcester Polytechnic Institute. Contact her at [email protected]. Dongning Luo is pursuing a PhD in computer science at the University of North Carolina at Charlotte. His research interests include data and visual analytics on largescale text collections. Luo has a BE in electrical engineering from Shanghai Jiao Tong University. He’s a student member of the IEEE Computer Society. Contact him at dluo2@ uncc.edu. Yujie Liu is pursuing a PhD in computer science at the University of North Carolina at Charlotte. Her research interests include visual analytics on large-scale text collections and evaluations. Liu has an MS in automatic control from Huazhong University of Science and Technology. She’s a student member of the IEEE Computer Society. Contact her at [email protected]. Selected CS articles and columns are also available for free at http://ComputingNow.computer.org.
EXECUTIVE STAFF PURPOSE: The IEEE Computer Society is the world’s largest association of computing professionals and is the leading provider of technical information in the field. MEMBERSHIP: Members receive the monthly magazine Computer, discounts, and opportunities to serve (all activities are led by volunteer members). Membership is open to all IEEE members, affiliate society members, and others interested in the computer field. COMPUTER SOCIETY WEB SITE: www.computer.org OMBUDSMAN: Email [email protected]. Next Board Meeting: 15–16 Nov. 2010, New Brunswick, NJ, USA
EXECUTIVE COMMITTEE President: James D. Isaak* President-Elect: Sorel Reisman;* Past President: Susan K. (Kathy) Land, CSDP;* VP, Standards Activities: Roger U. Fujii (1st VP);* Secretary: Jeffrey M. Voas (2nd VP);* VP, Educational Activities: Elizabeth L. Burd;* VP, Member & Geographic Activities: Sattupathu V. Sankaran;† VP, Publications: David Alan Grier;* VP, Professional Activities: James W. Moore;* VP, Technical & Conference Activities: John W. Walz;* Treasurer: Frank E. Ferrante;* 2010–2011 IEEE Division V Director: Michael R. Williams;† 2009–2010 IEEE Division VIII Director: Stephen L. Diamond;† 2010 IEEE Division VIII Director-Elect: Susan K. (Kathy) Land, CSDP;* Computer Editor in Chief: Carl K. Chang† *voting member of the Board of Governors
†nonvoting member of the Board of Governors
BOARD OF GOVERNORS Term Expiring 2010: Piere Bourque; André Ivanov; Phillip A. Laplante; Itaru Mimura; Jon G. Rokne; Christina M. Schober; Ann E.K. Sobel Term Expiring 2011: Elisa Bertino, George V. Cybenko, Ann DeMarle, David S. Ebert, David A. Grier, Hironori Kasahara, Steven L. Tanimoto Term Expiring 2012: Elizabeth L. Burd, Thomas M. Conte, Frank E. Ferrante, Jean-Luc Gaudiot, Luis Kun, James W. Moore, John W. Walz
Executive Director: Angela R. Burgess; Associate Executive Director; Director, Governance: Anne Marie Kelly; Director, Finance & Accounting: John Miller; Director, Membership Development: Violet S. Doan; Director, Products & Services: Evan Butterfield; Director, Sales & Marketing: Dick Price
COMPUTER SOCIETY OFFICES Washington, D.C.: 2001 L St., Ste. 700, Washington, D.C. 20036 Phone: +1 202 371 0101; Fax: +1 202 728 9614; Email: [email protected] Los Alamitos: 10662 Los Vaqueros Circle, Los Alamitos, CA 90720-1314 Phone: +1 714 821 8380; Email: [email protected] Membership & Publication Orders: Phone: +1 800 272 6657; Fax: +1 714 821 4641; Email: [email protected] Asia/Pacific: Watanabe Building, 1-4-2 Minami-Aoyama, Minato-ku, Tokyo 107-0062, Japan Phone: +81 3 3408 3118 • Fax: +81 3 3408 3553 Email: [email protected]
IEEE OFFICERS President: Pedro A. Ray; President-Elect: Moshe Kam; Past President: John R. Vig; Secretary: David G. Green; Treasurer: Peter W. Staecker; President, Standards Association Board of Governors: ; W. Charlston Adams; VP, Educational Activities: Tariq S. Durrani; VP, Membership & Geographic Activities: Barry L. Shoop; VP, Publication Services & Products: Jon G. Rokne; VP, Technical Activities: Roger D. Pollard; IEEE Division V Director: Michael R. Williams; IEEE Division VIII Director: Stephen L. Diamond; President, IEEE-USA: Evelyn H. Hirt
revised 17 Jun. 2010
IEEE Computer Graphics and Applications
41
