Web users are often overwhelmed by the amount of information available while carrying out browsing and searching tasks. Recommender systems substantially reduce the information overload by suggesting a list of similar documents that users might find interesting. However, generating these ranked lists requires an enormous amount of resources that often results in access latency. Caching frequently accessed data has been a useful technique for reducing stress on limited resources and improving response time. Traditional passive caching techniques, where the focus is on answering queries based on temporal locality or popularity, achieve a very limited performance gain. In this dissertation, we are proposing an ‘active caching’ technique for recommender systems as an extension of the caching model. In this approach estimation is used to generate an answer for queries whose results are not explicitly cached, where the estimation makes use of the partial order lists cached for related queries. By answering non-cached queries along with cached queries, the active caching system acts as a form of query processor and offers substantial improvement over traditional caching methodologies. Test results for several data sets and recommendation techniques show substantial improvement in the cache hit rate, byte hit rate and CPU costs, while achieving reasonable recall rates. To ameliorate the performance of proposed active caching solution, a shared neighbor similarity measure is introduced which improves the recall rates by eliminating the dependence on monotinicity in the partial order lists. Finally, a greedy balancing cache selection policy is also proposed to select most appropriate data objects for the cache that help to improve the cache hit rate and recall further.