Nov. 3, 2006 — Imagine tuning into your favorite news program and hearing: "On today’s newscast, we’ll talk about the war in Iraq, your high school football team’s victory, and a new music review from your friend Mary's website."
A news show that truly speaks to its viewers is the idea behind "News at Seven," an Internet-based program designed to automatically gather, edit and deliver a package of original news and information to people based on their specific interests.
Even the newscaster is personalized. The host is a computer-animated person broadcasting from a familiar location, such as your favorite beach.
"The entire environment can be customizable. We want it to be hyper-personalized," said Kristian Hammond, professor of computer science and director of the Intelligent Information Laboratory at Northwestern University.
At the heart of the program is software capable of grabbing a news feed from a website and using the text of the story to drive a search for images, video and opinionated blogs.
Specifically, the software picks up the most popular story under a certain category, such as "geopolitical" or "science" or "sports," for example.
Next, the software edits the story for size, so that the words correspond to a spoken segment 45 to 90 seconds long.
The software is capable of eliminating abbreviations and acronyms, converting any passive voice to active and modifying how quotes are presented so that spoken text seems conversational.
It then searches for particular entities mentioned in the story, such as names of countries, cities, people and buildings, for instance, and uses those names to drive a search for images and video.
The text is also tapped to search for blog postings that are related to the story. The software looks for phrases, such as "George Bush thinks," or "George Bush said," as well as emotionally charged words, such as "horrible" or "fantastic." It ranks the words as negative or positive then scores the words and phrases according to the level of emotion and picks out the highest scores.
The story is fed to a computer-animated character, whose tone and inflection are driven by language in the story's text and by the viewer's specific requests.
