As one of the key technologies to solve the information overload, the collaborative filtering algorithm exhibits the flaw of inaccuracy prediction. Therefore, based on the analysis of Spark technology as well as its framework and the elaboration of the flaw in Slope One algorithm，an improved Slope One algorithm has thus been proposed for the similarity between projects and users, followed by the implementation of the algorithm on Spark platform. Experimental results show that the improved Slope One algorithm has a higher accuracy of prediction with its paralleled implementation on Spark. The combined methods of Speedup and Sizeup prove that this algorithm is characterized with a good parallel effect and an excellent expansibility, thus helping to promote the efficiency.