Research And Implementation Of Collaborative Filtering Algorithm In E-commerce

With the rapid development of Internet, the scale of resources in E-commerce site raises sharply. Users hope to get information more timely while they are enjoying the convenience and efficient product information service from E-commerce. Users' requirement can be well satisfied by providing them personalized recommended systems which recommend users some resources they may be interested in by analyzing users'historical behavior. Existing recommendation technology can't match users' requirement for its lots of limitation, although it can resolve this problem in some way. This thesis mainly studied the theory of traditional collaborative filtering, proposed an improved algorithm against its scalability, and verified its feasibility by experiments. The main work in this thesis is as following:1. The status of the current recommendation system and classical recommendation technology were described. The basic methods and principle in traditional collaborative filtering were explored, and the nearest neighbor way was searched, as well as the advantages and disadvantages of different technology of collaborative filtering were analyzed.2. The limitation of traditional collaborative filtering was investigated from the aspect of recommendation principle, and the shortage in processing large-scale and high dimension data for traditional collaborative filtering was discussed.3. The Min-Hash based dimensional reduction method on large-scale data is studied, which was used in collaborative filtering. And then the collaborative filtering model based on Min-Hash was built and implemented.4. The comparative experiments were carried out on the given data. As for the data set with noise and synonymity, the solution for noise filtering and data preprocessing was proposed, which improved the performance of the method.Experimental results show that Min-Hash based method can reduce the computation cost both in time and space, and meanwhile ensure the recommendation quality, which can be applied in the large-scale E-commercial website.