Semantic models of news can enable richer interfaces for end-users to learn the context of news events referenced in blog posts. We present Brussell, a system that uses contentspecific models of news event situations to perform anticipatory information retrieval, organize extraction results and present a novel, structured interface for navigating among the events of a news situation.
Often background events are related to the event covered by a news article by being part of the same overall situation perhaps an earlier event in the situation, where by situation we mean a limited sequence of causally-related events, such as all of the newsworthy actions in a lawsuit. For example, an account of a lawsuit being settled by the defendant may refer back to the original filing of the lawsuit and both events are part of a particular lawsuit situation. Alternately, a news article may reference other similar or related situations. A similar lawsuit may be taking place simultaneously in another locale. Related lawsuits include a suit acting as a case precedent, or other suits involving some of the same participants, such as other suits against the defendant.
All of these relationships are part of the situational context that the user draws upon in making sense of the news event a news article describes. This context gives rise to specific questions, such as: • • • •
What happened in the other situations referenced in this article? What other similar and related situations have these participants been involved in? Neither conventional software and web sites for blogs and news, nor research systems for “semantic-blogging” provide content-specific support for answering these questions, however.
Without an in-page link, the user must find related articles manually in order to answer questions like these. She must identify relevant terms such as entity names and situation keywords. Then she cut-and-pastes them into a news search engine. Finally she sorts through lists of results to find relevant articles. These steps make for an inconvenient process familiar to anyone who reads news on the web. Even news timelines provided by advanced search engines are unable to provide content-specific overviews of a situation in accordance with the user’s expectations of how it begins and continues.
We present Brussell, a system that performs anticipatory information retrieval and model-based information extraction to support the user in exploring the situational context of the news. Brussell retrieves news articles and extracts information to create models of news situations. When a user selects a situation, it presents a storyline with the major milestone events. Clicking on the event label loads an article that either immediately covers the event or is the earliest mention of the event. Evidence that an event took place, for its date and location, or for important attributes of participating entities can also be viewed in the form of collected textual snippets and links to source pages. All of these features work together to provide the user a “big picture” view of the news situation and answer highlevel questions about it.
EXAMPLE Consider the case of a user reading about the kidnapping of a BBC journalist on a blog post. Although the user is vaguely aware of this incident, he would like to find out more. With standard search technology, he would enter terms into a search engine and peruse the results in order to develop an overall sense of how the kidnapping situation transpired. Through Brussell, he can interact directly with the text referencing the situation. Moving the mouse over the text causes it to be highlighted (see Figure 1). Right-clicking on this highlighted text opens a context menu presenting options for viewing the history of the situation and finding out more about its participants (see Figure 2). The user wants to see a summary of what happened, so he selects the first option, which updates the toolbar to show a storyline for the kidnapping with its major events and their dates (see Figure 3). With this high-level view of the kidnapping situation, he is able to see the how the situation unfolded through the occurrence of its milestone events. Alternately, selecting one of the menu items to learn more about a participant would load a similar timeline view presenting all of the situations the participant has been involved in.
Next, he wants to know exactly how the situation ended, so he selects the "release" event button that loads the most relevant page describing the event in detail (see Figure 4). By reading this page the user can learn more about the circumstances under which the journalist was released. ARCHITECTURE Brussell consists of a Firefox browser plugin and server software, which may both run on the same computer. When the user visits a web page the browser plugin sends the current page title and URL to the server, which responds with the (possibly cached) page situation references. A user can view situation references in a blog post, as in the example, or in other web pages such as news articles.
The back-end system uses manually created situation model
types (scripts) and currently supports kidnappings, legal
trials and corporate acquisitions each of which has multiple
possible outcomes and 8-12 possible events. The system
runs daily to retrieve news articles from several news web
sites via RSS feeds and store them in a Lucene index [
RELATED WORK
Several semantic-blogging systems and been developed to
support blog authoring for the Semantic Web. An extension
to the RDF-based personal information management system
Haystack supports right-clicking on any web page to create
a blog post with a link to the page [
Some systems extract formal knowledge from news to
populate the Semantic Web such as SemNews, which
processes news retrieved via RSS feeds [
Other work has focused on finding news articles relevant to
blog posts by identifying and searching for important words
in the post [