System Overview

The BINGO! system consists of six main components:

• the crawler itself,
• an HTML document analyzer that produces a feature vector for each document,
• the SVM classifier with its training data,
• the feature selection as a "noise-reduction" filter for the classifier,
• the link analysis module for determining topic-specific authorities and hubs, and
• the training module for the classifier that is invoked for periodic re-training.

BINGO! is implemented as a collection of Java servlets that run under the control of a Web application server. All intermediate data, such as "raw documents" that the crawler fetches or feature vectors (vectors of term weights), and also the final ontology index are stored in a database using Oracle8i. This way the various components and their execution threads are largely decoupled with regard to timing and performance. For example, the classifier is not affected by temporary delays in the crawler's threads as long as its input queue, a table in the database, is sufficiently filled.