The IRMA Community
Newsletters
Research IRM
Click a keyword to search titles using our InfoSci-OnDemand powered search:
|
Speedup Learning for Text Categorization and Intelligent Agents
Abstract
Research in text categorization has been focused on off-line machine learning algorithms: a predetermined set of categories is learned prior to the operation of the system in which they are to be applied. Also, the learning from examples paradigm requires a training session in which a teacher, rather than the enduser, manually labels the training set of example documents. This is labor intensive, particularly for rare categories: and essentially all categories on the Internet are rare. We propose the use of a speedup-learning algorithm in which a user interacts directly with the machine learning algorithm, and thereby greatly reduces the amount of training documents that must be labeled for optimal performance of the system. It also places the training capability directly into the hands of end-users, which opens up new applications, e.g. to track breaking news events on the Internet. Other researchers have previously identified the speedup learning strategy; we extend the concept, implement an algorithm, and apply it to Intelligent Internet Agents and law enforcement.
|
|