Semantic Based Web Agent for Web Prefetching

Abstract

Web caching and Web prefetching are the two most important Web latency toler- ance techniques that have been developed over the years. Though the caching gives a solution to this the bene t of caching is limited due to the rapid changes of network resources. As a complement to web caching, Web prefetching technique is to prefetch Web documents (Web objects) that would be accessed in the near future while a client is processing previously retrieved documents. Previous studies on web prefetching were based mostly on the server knowledge and URL graphs. In other words, servers keep track of clients' accesses to web objects. This historical information is represented by URL graphs. Most of these algorithms are e ective in prefetching but only a few can prefetch documents that are newly created or never visited before and consider about client's preferences. We have proposed an ontology-based semantic Web prefetching approach to over- come the limitations mentioned above. Semantic Web prefetching technique captures the user's searching interest from his/her searching pattern and past access patterns and predict future accesses from a list of possible documents when a new web site is visited. This approach is capable of prefetching objects whose URLs have never been accessed. Though this is not a novel approach in web prefetching, we discuss how to optimize the access predictions by building a user pro le and the challenges when extracting semantics from existing HTML pages. An agent application is implemented as a client- side prefetching prototype of the system. It monitors the user's web searching activities and keywords (concepts) that are found in user's accesses are used to build the user pro le. WordNet has been used as the ontology to together semantic distances among concepts. In the evaluation we concern about news services because they change their articles/documents very often and so do the links.