Researchers have studied query recommendation to address various aspects of the user search experience. Several contributions in this area use search logs to recommend existing queries from the log, using query-based similarity metrics or log-based probabilities. However log-based recommendations are limited to queries issued by users, not necessarily utilising the full potential of a search system. The proposed work here intends to approach query recommendation from a generative perspective. It proposes to generate novel queries by using search logs and web crawls to model a user’s knowledge and to recommend queries to satisfy knowledge deficiencies.