In order to reflect the actual case of users' decisions and rating patterns, and solve the sparsity problem of traditional collaborative filtering recommendation algorithms, a trapezoid fuzzy rating model is proposed, which fuzzifies crisp point into trapezoid fuzzy number based on the rating statistical information of users. The trapezoid fuzzy number contains personalized information of users and can reasonably represent user's preference degree and rating pattern. Based on this model, the user fuzzy similarity-based collaborative filtering recommendation algorithm is designed. We prove that this algorithm is a fuzzy extension of Cosine similarity, and analyze the application scopes of this method. The experimental results show that our method can obtain better performance than other traditional methods, especially when implemented in a sparse dataset with high ratio between users and items.