The Research On Personalized Recommendation Model And Key Technology Of Internet + Commerce

"Internet + Commerce" is a subversive innovation of the traditional e-commerce and it's future development will be more intelligent,service oriented and platform oriented.As a key technology for intelligence and service-orientation,personalized recommendation has been a research focus in the area of traditional electronic commerce.Therefore,it is particularly important to study the personalized recommendation of "Internet +Commerce".The appearance of "Internet + Commerce" brings new challenges to personalized recommendation technology.Combining with the individual requirements of"x-commerce-consumption-service-platform",this thesis solves some problems,such as modeling problem of traditional personalized recommendation about user's interest,the multi-interest recommendation and cold-start recommended problems.Besides,this thesis studies personalized recommendation of "Internet + Commerce" on the aspects of user interest model analysis and design,recommended algorithm design and implementation,recommendation framework and application.The main tasks are as follows:1.Propose user multi-interest model and user interest model based on interest label.Aiming at the problem that the single interest model can't describe the user's preference effectively,this thesis puts forward two kinds of user interest models.First,based on rating data and behavior data,the user multi-interestmodel(User-MI)model is proposed.This model gives the formal description of user multi-interest and illustrates it with examples;Then,according to the data's characteristics of the basic information of the platform,this thesis comes up with personas of interest benefit(POIL)model which is based on interest label.At the same time,it also designs an interactive user interest label updating algorithm which is focused on usere2.Design user multi-interest recommendation algorithm.First,this thesis gives a formal description to the problem and the problem is transformed into the construction scoring utility function and the construction multi-interest regular function;Then,based on the User-MI model,by limiting probability matrix decomposition methods,it establishes scoring utility function and multi-interest regular functions;Further,through the greedy approximation algorithm,it solves the objective function and makes comparative experiments.After the experiments,it proves the effectiveness of the algorithm for the multi-interest recommendation.3.Design interest label matching algorithm.Peer-peer relationships between user interest label and product's features are used in this research.And based on POIL model,this thesis investigates recommendation algorithm in view of interest label matching.By comparative experiments,this study verifies the validity of the algorithm for the cold start problem and summarizes the characteristics of the algorithm and its application.4.Implement the personalized recommendation application.This research provides an application framework of personalized recommendation.According to platform operation,this thesis analyzes the recommended functional structure.At the same time,it designs the use case diagram and class diagram and realizes the application of personalized recommendation service.