Conclusions

Abstract

This dissertation has described a scalable, efficient webcasting system that relieves the user from the burden of manually maintaining a personal interest profile. The central idea for automatically building interest profiles is to monitor the information that a user currently produces either by typing or by speaking. Discovering frequently appearing information patterns enables automated profile generation and allows for computerized updates to reflect stable interests. The critical challenge for a webcasting system with a large subscription database is to sustain tolerable delivery performance. This thesis addresses this challenge by combining a compact profile representation with dynamic data structures and efficient matching algorithms.