E-commerce Oriented Collaborative Filtering Recommendation Algorithm And The Recommendation System Research

With the rapid development of the Internet and the popularity of the PC and Table PC, electronic commerce has become an economic growth point in today’s IT and Internet industry. However, not only the number of users is increasing, at the same time, a large number of useful and useless information are in e-commerce. How to quickly find the information we need from thousands of massive amounts of information has become an important topic. In order to solve this problem, many personalized recommendation systems appear in e-commerce sites in recent years. Collaborative filtering algorithm is the most widely used, and the most successful recommendation technology. However, existing algorithms have not considered user’s interests changing with the time. When we recommend the products which have a strong dependent of the time, the recommendation quality can’t be guaranteed and, and the robustness is poor. At the same time, the cold start problem which appears in collaborative filtering algorithm will affect the recommendation quality. Based on the collaborative filtering algorithm and the knowledge-based recommendation algorithm, combining the property of the time, this article researches some recommendation algorithms’ performance. The main work is as follows:1) To find the user’s nearest neighbors, a similarity model based on cosine measure is established, and a traditional collaborative filtering algorithm is designed and implemented. We choose a different number of users set to test the algorithm’s performance. Experimental comparisons show that the recommendation accuracy rate is sensitive to the number of neighbor users, and it will obtain better performance when the number of neighbor users is selected appropriately.2) According to the time-dependent properties of the product, we propose a collaborative filtering algorithm based on time-weighted scheme. Based on the collaborative filtering algorithm, a time filter function is used to generate the recommendation. In this paper, we design two kinds of time-weighted functions. Furthermore, in the recommend procedure, different weights are given to the different items according to the time-dependent properties and the actual time of the evaluated item. Experimental results show that the collaborative filtering algorithm with time filtering function can quickly discover the changes of the user’s interests, and recommend products with better performance to meet personal requirements of different users.3) In order to discover the changes of the user’s interests and overcome the cold start problem that appears in traditional collaborative filtering algorithm, we present a hybrid recommendation algorithm. Combining a time filter function with a knowledge-based recommendation procedure, we put forward a fusion of time algorithm, which can be viewed as a hybrid algorithm with knowledge-based reasoning and collaborative filtering mechanism. This algorithm can effectively overcome the cold start problem, that is, when the number of user evaluations is small, the algorithm uses the knowledge-based reasoning to generate recommendation, when the number of user evaluations reaches a certain level, the algorithm uses the collaborative filtering mechanism to generate recommendation.4) According to the proposed hybrid recommend algorithm which combines knowledge-based reasoning with collaborative filtering mechanism, we design a movie recommendation system which can be used in e-commerce. The system can search the nearest neighbors according to the movie scores rated by users, and give the user the best movies recommendation.