This paper develops a Semantic Web application, namely WorldWideWatch or WWWatch, in short. By employment of Semantic Web technologies, WWWatch analyses and visualizes social networks between countries in the world from information extracted from online news archives. Such information is described using RDF/OWL ontology and stored in Sesame server. The social networks in this application are colored and shaded in order to represent relation polarities (i.e. positive, neutral and negative) and the confident levels of relationships among countries. Users can query social relationships of certain countries as well as related events occurred in the specified time period and display the results in different forms such as graphs and timeline diagrams. The polarities of relations are taken into account in the analysis of social networks. In addition, the proposed system can display the location of some world phenomena such as natural disasters and manifestations of global warming on a map with respect to user-specified time periods, regions and phenomenon.
Existing social network applications [4, 6] mainly focus on structural information (e.g. centrality, closeness, betweenness, etc.) of unsigned binary graphs or valued graphs. This implies that the analysis only considers whether there exist any relations among subjects. However, in the real world, social relationships among subjects (such as people, animals, organizations, parties, countries etc.) can be more expressive.
A relation between two subjects is positive when it is pleasant to both of them whereas a neutral relation implies that the subjects do not show what they are thinking about relationship. Alternatively, the neutral relation can mean no relationship. On the other hand, whenever subjects feel upset about each other, their relation will become negative. These introduce an attribute for social relationships, called Relation Polarity, which in nature changes over time. Another interesting attribute is Relation Confident Level. Since social relationships are implicit knowledge extracted from explicit documents. It is important to quantify how much a piece of information contributes to the relationship between a pair of subjects. Furthermore, it also depends on how many documents exist in the specified period of time.
To demonstrate the importance of these two attributes in social network analysis, WWWatch, a Semantic Web application, is developed in this work and the online prototype system can be found at: http://ict.shinawatra.ac.th:8080/wwwatch. WWWatch enables users to visualize social networks of countries in the world based on information extracted from various online news archives. Colors and shades are employed to illustrate the relation polarity and the relation confident level, respectively. The relation polarity is determined based on reference emotion dictionaries. Users are able to query by countries and time period. They can view resultant social networks as either graph or timeline diagram along with the analysis. It is important to emphasize that the social networks shown in the application are for demonstration purposes only. To reflect actual relations, significant amount of reliable documents and more comprehensive emotion-extraction techniques are required.
In addition, the proposed application can display the location of world phenomena such as disasters and manifestations of global warming on a map with respect to user-specified time periods, regions and phenomenon. The information about the occurrences of phenomenon is also gathered from the online news archives. This demonstration aims to warn people about the climate changes and motivate a movement to treat the world better by means of Semantic Web technologies.
The paper is organized as follows: Section 2 introduces WWWatch System via a running example. Section 3 describes the proposed system architecture, details its components and discusses implementation issues. Finally, the conclusions and direction for future research are presented. 2
This section demonstrates WWWatch by means of an example. Suppose that a user
wants to retrieve social relationships between Thailand and Singapore of the year
199
Colors are used to represent relation polarity, i.e., edges in green means a positive sign, while those in red implies a negative one. Moreover, the system can show the relation confident level using color shading. The darker the edge color, the stronger the relationships; on the other hand, the lighter the color, the weaker the relationships. A user can additionally view labels of edge weights for more precise information. Such exploration and visualization can be part of a sociological study. Note that section 3 discusses more details of the edge analysis.
The timeline diagram aims to reflect the relationships of two specified countries
over the past ten years (199
In general, any online Web documents with URIs can be used as knowledge resources for the proposed application. In this prototype, two different types of resources are used: HTML pages from Google News Archive [10] and RSS pages from Yahoo! News and Google News [10, 15]. For Google News Archive, a few of Web mining techniques are employed whereas SAX Parser is used for parsing RSS files. Extracted information is described in RDF/OWL ontology and stored in SESAME [3, 13] database.
The RDF database comprises news items which are related to the following 10 countries: Afghanistan, Australia, China, France, Japan, Iran, Thailand, Singapore, South Korea, and the United States. For each pair of these countries, top 100 news results that both countries co-occurred in each year between 1997 and 2007 are collected. Another dataset is created by collecting top 100 news results that a natural phenomenon and the specified country name co-occurred. In this application, nine types of natural phenomena are selected, such as, earth quake, tsunami, wildfire, heat wave, hurricane, drought, flood, disease spreading, and high-temperature weather. This layer concerns geographic lookup, temporal lookup, and relationship reasoning and social network clustering. This section mainly focuses analyzing collected data to determine relation polarity and relation confident level between countries. Relation polarity aims to measure an emotional valence hidden in the existing relationship whereas relation confident level indicates how much a user can be confident about the discovered relationship. Both attributes can be measured by Relation Confident Level with Polarity (RCLP) which will be explained in this section.
In this work, emotional valence of a word is simply categorized into 3 types:
positive (+), negative (-) and neutral (0). In order to classify emotional valence of a
news article, emotional dictionaries are a mandatory component. A few research
works in emotional text analysis have been proposed [1, 5, 7, and 8] and they built
their own emotional dictionaries. However, none of them provides a
publicaccessible emotional dictionary. Therefore, in this work, an emotional literacy
database has been created by extracting information from the following sources:
! Emotional Literacy and Dictionary
To analyze the emotional valence score of a news article, partial keyword matching algorithm is employed to match emotional words in the dictionary with the news title and description. By default, the article is assumed to be neutral (0). If it contains positive words more than negative words, its cumulative emotionalValenceScore will be greater than zero which implies that this news sounds positive (+1). On the other hand, if it has more negative words, its cumulative emotionalValenceScore will be less than zero and its polarity is -1. In addition, if the numbers of positive words and negative words are tie or none of these words appears in the news article, the news polarity remains zero.
Since the relationship is computed from any Web resources, the quality of information becomes crucial to indicate the reliability of such relationship. Hence, in this application, the number of related news articles to the focused news article is used as the key quality measure. Thus, the relation confident level with polarity can be computed as follow: Let NDB(p,q) = {n1,n2,…,nm} be the news collection comprising m news articles that contain both countries p and q and R(ni) be the number of related news articles of a news article ni. The relation confident level with polarity between the countries p and q with respect to an emotional dictionary EDB is: m $ emotionalValenceScore(ni , EDB) * R(ni ) RCLP( p, q, EDB) # i#1 , where ni " NDB( p, q) m $ R(ni ) i#1
The confident level of a relation is always |RCPL|. However, if RCLP > 0, it indicates that the relation between p and q is positive whereas if RCPL < 0, it implies negative relation. In addition, RCPL = 0 means neutral relationship or no tie between these two countries. 3.3
With an emphasis on ease of use and installation, WWWatch system is designed to be implemented as a Java Web application based on the client-server model. The clientside employs Java Applets to create the interaction part embedded in HTML pages for users browsing. On the server-side, a Java Server Page (JSP) and servlets are used to create the system applications that hide the complexity of services. Furthermore, the application exploits a variety of Semantic Web technologies to provide various graphical models as follow: ! Country Social Network: In order to represent a country network model as a graph, the system employs the service of JUNG (JAVA Universal Network/Graph) [9] programming toolkit, a JAVA based software library developed for data manipulation, analysis, and visualization. Relation Confident Level with Polarity (RCLP) is used to calculate weights for edges in the graph and their color shades as shown earlier in Fig. 1. Users are allowed to query data by country and year. ! Timeline of Relationship: Simile’s Timeline [14] is selected as a tool to visualize news articles related to two specified countries over the past ten years. It is a powerful DHTML-based AJAX widget for creating dynamic explorable timeline. To illustrate emotional valence of news and relationships, background color of the timeline and colored bullets are used. In this interface, users have to provide two country names and year. Since only the news headlines are shown, to see more description, users may click on its bullet. Due to large number of news items, viewing timeline of relationship on the Web can be significantly slow. ! World Phenomena Geographic: An open source JavaBeans toolkit, namely BBN’s OpenMap [12], is employed in this visualization. Several useful features of such toolkit enable users to add, remove and change the positioning of different components and layers and to freely navigate over the map. However, it consumes a significant number of resources and may take several minutes to display over the Internet. 4
WWWatch is a Semantic Web application designed for visualizing social networks of countries in the world from information extracted from online news archives. To illustrate the polarity and confident level of relations among countries, colors and shades are used and users are able to search and display results in various formats such as graphs and timeline diagrams. In addition, the proposed system can display the locations of natural phenomena on a map based on user-specified periods and disaster types. WWWatch development is in progress with continuing enhancement. Currently the prototype system’s database comprises data of only 10 selected countries; collection of more countries in the world is under way. More features will be added to the system in order to improve the quality of results and facilitate user search activities such as adding more criteria for user to query relationships and phenomena, optimizing the applet to speed up downloading time, improving emotional dictionaries and news mining techniques to increase precision and providing more results from social network analysis.