Using document features to optimize web cache

Abstract

In this paper Web cache optimization using document features is proposed. The problem in Web cache optimization is to decide which strategy to use in replacement of cache objects. While commonly used policies use heuristic rules, proposed model predicts the value of each Web object by using features collected from the HTTP responses and from the HTML structure of the document. In a case study, generalized linear model and multilayer perceptron committee model are used to classify about 50000 Web documents according to their popularity. Results show that linear model does not find any correlation between the features and document popularity. MLP model gives better results, yielding mean classification percentages of 64 and 74 for the documents to be left or to be removed from the Web cache, respectively.