This work focused on recommended accuracy of the collaborative filtering system to process large-scale sparse data. The cosine similarity measurement and correlation similarity measurement could not handle the sparse rating matrix very well. And these methods undermined the recommended system performance when dealing with large-scale data. This paper provided a new approach based on the similarity of rating interval. The user rating range was divided into several qualitative intervals. According to the rating ratios

user preferences and item qualities were represented by vectors. Finally

this work used the cosine similarity method to determine the nearest neighbors. This approach reduces the impact of the sparse matrix and has the ability to deal with massive data. On MovieLens data set

experiment result shows that the algorithm has reliable and accurate performance.