The rapid development of the Internet has led us into the era of network economy and information society, which had a profound impact on enterprises and personal lives. Internet-based e-business companies no longer need the physical connection to its customers and suppliers like a traditional corporate entities does. The emerging user-centric Internet sites also break lots of limitations of traditional media and spread their influence to a broader range of potential customers.In such circumstance, the number of goods or information network applications can provide is huge, making the users unable to just glance through the screen and find the desired information. Therefore, service providers need some intelligent algorithms, which would recommend goods or information to a user based on knowledge like his or her hobbies, in order to enable users to access what they want easily. And from a practical experience, user requirements are often vague, unclear, and may be of particular goods or information with the existence of potential preferences. If the service provider can fit the recommendation to the user, it may bring the user's potential demand into reality. In this context, recommendation systems came into being, it recommend possible objects that meet the user requirements based on the user's characteristics. Examples include e-commerce personalized recommendation system, or socialized news recommendation system, etc.Recommendation systems can be implemented in a number of ways, but because the data it faces vary widely, there is no recommendation algorithm that can perform well under all circumstances, and real applications usually employ a mixed recommendation strategy. In this paper, we examined several elementary recommendation algorithms and application scenarios, and proposed a new recommendation algorithm. It is prototyped on association rules based recommendation system and employs rank correlation to change the method of calculating degree of support fundamentally to overcome the shortcoming of data discretization in traditional association rules algorithm, so as to improve recommendation accuracy. Then, we built a multi-function, multi-algorithm personalized recommendation system prototype based on the algorithm we proposed and some classic ones such as collaborative filtering. Finally, the prototype system is applied on a classic dataset of recommendation system research, by comparing the accuracy of different algorithms' results, algorithm presented in this paper showed a good practicability. |