Web caching is a well-known strategy for improving the performance of web systems. The key to better web caching performance is an efficient replacing policy that keeps in the cache popular documents and replaces rarely used ones. When coupled with web log mining, the replacing policy can more accurately decide which documents should be cached. In this paper, we present a PLSA based prediction model to predict the user access patterns and interest to extend the well-known NGRAM-GDSF caching policy. Extensive experiments are conducted on the publicly available web logs datasets. The result shows that our approach gets better web-access performance.