In order to improve the accuracy of user similarity calculation and knowledge push, this paper puts forward a collaborative filtering push algorithm based on hot item punishment and user interest change. Firstly, this algorithm is utilized to cluster knowledge items into several classes; and then in each class, the user interest degree function is introduced to predict the rating value of unrated knowledge items; and in user similarity calculation of each class, hot item weight coefficient is introduced to punish the influence of hot items on user similarity; and finally weight coefficient of user interest change with time is introduced to push. The experiment also uses Movie Lens data set to test the algorithm, and the results show that the improved algorithm is better than traditional collaborative filtering algorithm in terms of push accuracy.