The collaborative filtering technique is widely used in recommendation systems across a variety of fields. However, it faces problems such as data sparsity and user interest drift, making it difficult to generate high-quality recommendations. In this paper, we use a Fuzzy C-Means clustering algorithm that combines item attributes to cluster items, add a time weight factor to the similarity calculation based on the Ebbinghaus forgetting function, and then find the nearest neighbor set by applying an improved similarity formula to generate recommendation results. The empirical results show that: (1) the Fuzzy C-Means clustering algorithm used in this paper that combines item attributes can effectively improve the recommendation results, and is able to find a more accurate nearest neighbor set compared to traditional methods; (2) The improved time weight function based on the Ebbinghaus forgetting function can adjust the weights of items at different times, which can effectively improve the problem of user interest drift.