model-based methods in [2] However, the multidimensional approach described in [4] and the classical two-dimensional recommendation methods have one significant limitation in common. These methods are hard-wired by the developers into the recommender systems, are inflexible and limited in their expressiveness, and therefore, neglect some possible needs of the users. For example, a typical recommender system would recommend the top k items to a user, or the best k users for a product. This situation is quite limited especially in multidimensional settings, where the number of possible recommendations increases significantly with the number of dimensions [5]. Therefore, there is a need to empower end-users and other stakeholders by providing them with the tools for expressing recommendations that are of interest to them [3, 15]. For example, Jane Doe may need a recommendation for the best two dates to go on vacation to Jamaica with her boy friend. Also, Netflix or an on-demand movie service, such as provided by the Time warner Cable, can envision a web-based interface to a multidimensional cube of ratings that lets the users express the recommendations that are of interest to them or automatically tailors recommendations based on a given context, such as the time of day or the day of week. For example, a certain user(. g, Tom) may seek recommendations for him and his girlfriend of top 3 movies and the best times to see them over the weekend, and he enters this request into the recommender system via the web-based interface. Such query-based recommendation applications are not limited to on-demand movies but are relevant to a broad range of recommendation applications, including retailing, financial, travel and other applications. Furthermore, we believe that flexible recommendation capabilities would be appealing to a variety of different users, and not just to the end-users who are direct recipients of recommendations. For example, such functionality would be useful to the analysts of a company providing recommendation services, who may want to take advantage of all the knowledge that their recommender system holds and analyze it from a variety of different perspectives("show me the top 2 movie genres for each user age bracket", etc. One tool for expressing such requests is a recommendation language that is similar to how database sers use query languages to retrieve information from databases. In fact, one may try to use SQL for this3 model-based methods in [2]. However, the multidimensional approach described in [4] and the classical two-dimensional recommendation methods have one significant limitation in common. These methods are hard-wired by the developers into the recommender systems, are inflexible and limited in their expressiveness, and, therefore, neglect some possible needs of the users. For example, a typical recommender system would recommend the top k items to a user, or the best k users for a product. This situation is quite limited, especially in multidimensional settings, where the number of possible recommendations increases significantly with the number of dimensions [5]. Therefore, there is a need to empower end-users and other stakeholders by providing them with the tools for expressing recommendations that are of interest to them [3, 15]. For example, Jane Doe may need a recommendation for the best two dates to go on vacation to Jamaica with her boyfriend. Also, Netflix or an on-demand movie service, such as provided by the Time Warner Cable, can envision a web-based interface to a multidimensional cube of ratings that lets the users express the recommendations that are of interest to them or automatically tailors recommendations based on a given context, such as the time of day or the day of week. For example, a certain user (e.g., Tom) may seek recommendations for him and his girlfriend of top 3 movies and the best times to see them over the weekend, and he enters this request into the recommender system via the web-based interface. Such query-based recommendation applications are not limited to on-demand movies but are relevant to a broad range of recommendation applications, including retailing, financial, travel and other applications. Furthermore, we believe that flexible recommendation capabilities would be appealing to a variety of different users, and not just to the end-users who are direct recipients of recommendations. For example, such functionality would be useful to the analysts of a company providing recommendation services, who may want to take advantage of all the knowledge that their recommender system holds and analyze it from a variety of different perspectives (“show me the top 2 movie genres for each user age bracket”, etc.). One tool for expressing such requests is a recommendation language that is similar to how database users use query languages to retrieve information from databases. In fact, one may try to use SQL for this