Research On Electronic Commerce Recommendation Under Double Trust Mechanism

With the rapid development of Internet technology and the growing popularity of smart devices,e-commerce concept deeply rooted in people's minds.E-commerce has brought convenience to people,but also facing a serious problem of information overload.It's difficult for users to find what they want in the face of massive information,even putting in a lot of time and effort,then e-commerce recommended system followed out.Currently recommended system in e-commerce has been widely used,on the one hand to be able to provide users with convenient precise recommendation,on the other hand can increase the income of the business.Collaborative filtering technology is one of the most commonly used technology in the electronic commerce recommendation system,however,in the application process,there are problems such as sparsity,cold start and poor accuracy.The introduction of trust mechanism can effectively alleviate the above problems.In this paper,the shortcomings of traditional collaborative filtering technology have been improved,first we join the jaccard similarity in the calculation of similarity and direct trust degree,considering the proportion between common score and total project,solving the problem of the total number of two users giving a mark to only one project and calculated similarity is 1 in traditional similarity measure.Then,in the calculation of the trust model,the concept of double trust mechanism including direct trust and implicit trust is added which can better characterize the link between the user and the user.In the process of trust measurement,direct trust is calculated by mean squared difference and jaccard similarity,implicit trust is calculated through the propagation and aggregation of trust,implicit trust is added to contact the original customers which are not directly contacted,it makes the original sparse similarity matrix more denseness and improves the accuracy of trust measurement.Next,we merge the trust and similarity into a comprehensive weight applied to the prediction score.Finally,based on the classic film score movielens data set,we do a series of experiments.By comparing the user-item rating matrix?the direct trust matrix and the double trust matrix verifies the advantage of the presented algorithm to solve the problem of sparsity;By comparing the coverage rate between traditional collaborative filtering algorithm and the algorithm this paper represent verifies the later can get more recommendations;By comparing neighbor users between introduce the jaccard similarity or not verifies accuracy after joining jaccard similarity;By comparing and analyzing the MAE value of the three algorithms verifies the accuracy of the proposed algorithm.The results of the analysis can be found that the proposed model is superior to the traditional collaborative filtering algorithm,and will be well used in e-commerce recommendation system.