A method and system recommends items such as products and services to consumers. Rating on items are received from consumers as a sequential stream of data. A thin singular value decomposition is updated, one rating at the time, while receiving the ratings. A prediction of a recommendations of particular items for a particular consumer is based on the updated singular value decomposition while receiving the ratings and updating the singular value decomposition. The ratings are discarded after the updating so that a size, structure, and content of an underlying preference matrix is unknown.