Recommender systems typically generate recommendations for a user based on their profile, or for an item given its user interactions, but there are many scenarios especially in leisure domains such as books, movies, games and music, where users have specific recommendation needs, where they want to steer the recommendation process towards certain aspects they find relevant. Currently, there are few recommender or search systems that can deal with the complexity of such directed needs, nor do we know well which data types (metadata, user ratings and reviews, item content) are useful to match against different aspects of recommendation needs. There are many discussion forums where users describe their needs and their frustration with current search and recommender systems. In this talk I will summarize our work on analyzing relevance aspects for these needs and describe experiments on dealing with these.