Collaborative filtering is one of the most successful personalized recommendation techniques currently used in E-commerce recommendation systems. However, traditional collaborative filtering algorithms assume  the ratings to be static in each period. To solve this problem, two kinds of fuzzy cognition are proposed: fuzzy increasing of ratings and fuzzy increasing of time weights. Firstly, item ratings are divided into time windows, the similarity between items is calculated using a chain-structure, and the nearest neighbors of the target item is selected. Secondly, time weights are assigned to ratings, a weight function is proposed, and the traditional prediction method is improved. At the same time, a hierarchical optimization strategy is proposed in the prediction phase to solve the time weights of ratings, thus completing the recommendation. Finally, experiments on the Netflix datasets show that, compared with the traditional collaborative filtering algorithms, our proposal improves the recommendation accuracy by 9.8%~14.1%.