The user interests models in the existing recommendation systems can express the user’s interests effectively

but they can’t update the user interests models in time while the user interests change. This paper proposes a recommendation system of messages depending on user’s feedback to modify user interests. It can update model and the feature vector in time

and then the most matching recommendation results for every user. In addition

the system can better adapt to the growth of the scale of data by using HBase（Hadoop Database） as storage.