The given approach refines the traditional Static-Dynamic Cache (SDC) to capture the temporal locality of the topic in query streams. In SDC, static cache is populated periodically with the popular queries with assumption being popular queries in past remains popular in future. The dynamic part of the cache using LRU strategy captures the burst of short-term popular queries. The drawback of this approach is that it fails to capture queries with long-term popularity. There are certain queries which are neither too popular to be cached in static cache nor too frequent in short-term to be captured by dynamic cache. For instance, queries for certain topics like weather can be popular in the morning and late in the evening but are not popular in other parts of the day. To capture this temporal locality of the topic, topic cache is introduced along with the static and the dynamic part.

Assigning topic to a query is a challenging part as queries are generally very short. To classify queries correctly, queries are enriched with the search results. For this deliverable, queries are not enriched with search results, instead a news dataset is used to train the Latent Dirichlet Allocation (LDA) model. LDA model is an unsupervised statistical method to discover the hidden topic in the text. Given, k different topic, it assigns individual scores between [0, 1] to each topic. The topic is assigned to the query if the topic has maximum score and is greater than minimum threshold.