A Research Of E-marketing Recommendation System Based On Forgetting Curve And Domain Nearest Neighbor

With the rapid popularization of the Internet and the continuing prevalence of e-marketing, e-commerce system provides unlimited display space for businesses to offer customers more choices of goods, while its structure is also becoming increasingly complex. On one hand, customers feel helpless in face of a flood of product information and cannot successfully find their required goods. Problems like "information loss" and "information overload" have become more serious. On the other hand, businesses even cannot successfully get contact with customers, not to mention how to take sales measures. The emergence and application of e-marketing recommendation system has greatly alleviated such range of issues, which can effectively retain customers and prevent losing customers and improve the number of cross-sales of e-commerce system. However, the current e-marketing recommendation system is immature in actual application, it has serious problem of data sparsity and cannot successfully deal with the problem that customers who change their interests, such situation may causes serious problems like poor quality of recommendation and poor instantaneity of recommendation.For the challenges faced by the e-marketing recommendation system, this paper analyzes the research status of the current recommendation system and main advantages and disadvantages of recommendation technologies, this article also introduces a e-marketing recommendation method of calculation based on the existing hybrid recommendation studies, which is based on collaborative filtering as the main function and content-based filtering as the auxiliary function. Moreover, the method establishes user interest model based on "non-linear gradual forgetting curve" to predict the unevaluated rating of non-target users; to solve problems of data sparsity and users’interests drifting; then introduce the "domain nearest neighbor" way to find the nearest neighbors of target users to predict the unevaluated rating, so as to make recommendation based on this to effectively enhance the recommendation quality and efficiency of the recommendation system. This improving method fully uses the users’demographic information to calculate the similarity among users, making the recommendation results more in line with users’needs. The main research work of this paper includes:(1) through the analysis of the current situation of recommendation system domestic and abroad, it explores widespread problems such as the lack of personalized recommendations, single recommended method and low level of automated recommendation;(2) it proposes the hybrid recommendation method based on forgetting curve and domain nearest neighbour method, describes each improved procedure and calculation step of the proposed method in detail;(3) based on the public data sets of Movielens and by comparison pilots, it compares the recommendation effect between the e-marketing recommendation method proposed in this paper and other traditional recommendation methods to obtain objective and effective test results.