Web access latency reduction using CRF-based predictive caching

Reducing the Web access latency perceived by a Web user has become a problem of interest. Web prefetching and caching are two effective techniques that can be used together to reduce the access latency problem on the Internet. Because the success of Web prefetching mainly relies on the prediction accuracy of prediction methods, in this paper we employ a powerful sequential learning model, Conditional Random Fields (CRFs), to improve the Web page prediction accuracy for Web prefetching. We also propose a predictive caching scheme by incorporating CRF-based Web prefetching and caching together to reduce the perceived waiting time of Web users further. We show in our experiments that by using CRF-based Web predictive caching, we can achieve higher cache hit ratio and thus reduce more access latency with less extra transmission cost when compared with the predictive caching methods based on the well known Markov Chain models.