The University Record, April 8, 2002

Our top story: ‘Newsinessence’

By Jay Jackson
School of Information

No time to read the morning paper? When you search a particular news story on the Internet, do you get too much information?

A team of U-M researchers has developed NewsInEssence, a free, Web-based service that automates the collection of multiple, related news stories concerning a given event. These multidocument summaries are created for a user’s topic of interest and are automatically refreshed as news sources provide updates.

For example, users could choose to follow news about the war in Afghanistan or the latest stories about a corporate merger.

Dragomir Radev, assistant professor of information, and of electrical engineering and computer science, is heading the ongoing research project with graduate students from the School of Information and the College of Engineering.

NewsInEssence ( www.newsinessence.com) pulls news stories from CNN, the BBC, MSNBC, USA Today, Yahoo, and others. A feature that makes NewsInEssence different from other news-gathering services is that users can specify the length and frequency of news summaries.

“In comparison to other similar systems on the Web, NewsInEssence is personalized and interactive,” Radev explains. “It allows users to specify a ‘seed story’ or topic, to specify which sources they prefer and how long the search should take, to say how long the summary should be, and to decide how often updates should be sent to them.”

After NewsInEssence summarizes the entire set of documents it finds, users can change summary parameters, if they wish, and re-run the summary. NewsInEssence can also send summary updates directly to a user’s e-mail account at times the user chooses.

“The value in having NewsInEssence,” Radev points out, “is that news readers can quickly gather information about a given news story from multiple sources, thus providing them with different perspectives on the same news topic.”

Radev notes that NewsInEssence is a work in progress and an example of how students—while contributing their expertise as they learn new concepts in information search and retrieval, computer science, and linguistics—can provide a valuable public service. Soon to be added is a Web-based, natural language search query that will allow users to ask a question like, “When was the president of France inaugurated?” and get a concise answer.

The National Science Foundation Information Technology Research Program provided partial funding. The bulk of the programming was carried out at U-M by the multidisciplinary Computational Linguistics and Information Retrieval Group. Adam Winkel of the College of Engineering; Michael Topper of the College of Literature, Science, and the Arts; and Sasha Blair-Goldensohn, now a graduate of the School of Information, were the primary student programmers.

The backend summarization software, MEAD, was developed under Radev’s direction at a workshop last summer at Johns Hopkins University and has been placed in the public domain.