A. Georgakis (Sweden)
Prefetching, user behavior modeling, bigrams
In this paper a client-side algorithm that learns and predicts user requests is presented. The proposed ap proach is based on a user behavior profile. The profile is based on textual information extracted from visited web pages. The novelty of the paper is in the use of a part-of-speech tagger to filter the useful user-keywords. The keywords comprising the profile are employed by a transparent and speculative link weighting mechanism. The generated weights are used in estimating future web traversing. Afterwards some linked web pages are prefetched and stored locally in the browser’s cache. A comparison between the proposed algorithms and four other client-side algorithms yield improved cache-hit rates given a moderate bandwidth overhead.
Important Links:
Go Back
