Hybrid filtering of web pages for a recommendation agent

This article presents a web agent called BookmarkFeeder that recommends web pages for a user. It learns user interests by reading the user's bookmark items and monitoring user behaviour. Based on the automatically constructed user profile, it collects and filters web pages to recommend related web pages as bookmark items. BookmarkFeeder has two features that distinguish it from other web agents. First, it uses implicit feedback for learning. It learns the user's interests by monitoring the user behaviour on the bookmark items without relying on explicit feedback. Second, it uses a hybrid filtering strategy that uses link-based recommendation in conjunction with content-based recommendation. By using the hyperlink information, a web page that contains no text can also be recommended. The performance of the proposed system is demonstrated through experiments with untrained users.