Making Computers Smarter
Researchers at the Haifa, Israel-based Technion-Israel Institute of Technology have discovered a method for giving computers the type of knowledge that can make them think in a human-like manner which includes making logical connections between topics. This new type of computer knowledge can enable the machines to better filter junk emails (spam), make more efficient searches of the information highway, and even gather intelligence in a more sophisticated manner than previous computer programs. This is according to researchers Evgeniy Gabrilovich and Shaul Markovitch of the Technion Faculty of Computer Science.
Providing A Context
Researchers used a concepts database they created with online encyclopedia Wikipedia as their prototype. The English language section of Wikipedia contains more than one million articles alone. These concepts provide a context in the form of background material for computers struggling with web searches, for example.
Helping computers utilize this type of knowledge has long been a problem in the field of artificial intelligence, said Markovitch. "Humans use a significant amount of background knowledge" to understand text, "but we didn't know how to have computers access such knowledge," he said.
Computers can calculate how many times specific words will occur within two texts, and this helps them look smarter than they are within the parameters of a web search or an email filtering program, said Markovitch, who explained that the computers don't have any actual understanding of the text. The machine can only see the words as a collection of matching symbols, and has no real understanding of the sense of the words themselves.
Markovitch gave an example of how this works. An email filter may be configured to block messages containing the word "vitamin," but emails containing the word "B12" will continue to flow into one's inbox. The program has no inkling that B12 is a type of vitamin. "With our methodology, however, the computer will use its Wikipedia-based knowledge base to infer that "B12" is strongly associated with the concept of vitamins, and will correctly identify the message as spam," he added.
The Technion team is hoping to help computers make connections between disparate texts that share concepts. For instance, a section of text which refers to both Saddam Hussein and weapons of mass destruction should be able to be seen by computers as related to articles in which the words "war in Iraq," or, "U.S. Senate debates on intelligence," appear, even though none of these words can be found in the first text exemplar.
The database of concepts also aims to teach computers to distinguish between amorphous terms like "mouse" which sometimes refers to an animal and at other times to a computer device. The Technion is in the process of obtaining a patent for their work and believe their methods will be of great interest to both the intelligence community and web search engines.