In an approach for eliciting user demands for item recommendation, one or more computer processors retrieve one or more items based on a user demand. The one or more computer processors update the one or more items based on the user demand. The one or more computer processors extract the one or more representative words corresponding to the one or more items. The one or more computer processors build a candidate item list based on the one or more representative words. The one or more computer processors generate one or more eliciting questions to help a user select an item based on the candidate item list.