A Predictive Model for Web Prefetching

A prefetching strategy can help individuals more effectively to reduce the perceived access delay in Web information retrieval. Some prefetching algorithms have been proposed in literatures, but previous results show the ability is limited. This paper proposes an approach on predictive prefetching, which consists of an adaptive session generation method to extract navigational behaviors from Web log and a session-tree framework to organize sessions for fast retrieval. We first present a method to segment historical information of individuals' browsing behaviors into sessions based on time gap and session similarity. Session-tree is then built from these sessions. It enables us to organize historical information easily and retrieve subsequences both by events and timestamps efficiently. On the basis of these mechanisms, we predict individuals' coming requests and prefetch some results. Further, the experimental comparison with the standard PPM algorithm is also present. Through analyzing experimental results based on a real stock data set, we conclude the performance based on the adaptive session generation and the session-tree framework can predict well.