Features

By Robert S. Boyd, Knight Ridder Newspapers, Feb 1, 2006

WASHINGTON - ... The programs scour the data for hidden patterns or relationships, such as a suspicious number of insurance claims by an individual or repeated phone calls between, for example, Afghanistan and Detroit.

Data mining turns up such potentially meaningful patterns as, say, Person A telephoned B, who e-mailed C, who met with D and E, who rented an apartment together in F-town. Someone at that apartment made a phone call to someone in Country G in the Middle East. Human investigators can take it from there.

Jeffrey Ullman, a computer scientist who teaches a course on data mining at Stanford University in Palo Alto, Calif., offered a hypothetical example: Suppose you wanted to check a list of 10 suspected evildoers to see if any two of them spent two nights in the same hotel at the same time, perhaps to plot a terrorist attack.

According to Ullman, you'd have to search through at least 250,000 names to spot the suspicious meeting. That's too much for a human analyst but not for a computer. Here is the rest of the story.