| With the rapid development of the Internet industry, convenient and efficient e-commerce has become an indispensable part of people’s daily life. In the field of advertising, Internet advertising alliance has also gained considerable development. As one of the core technologies, personalized advertising recommendation is gradually becoming the key competitive technology between advertising alliances. Personalized recommendation in the advertising alliance refers to the application of technologies such as content filtering, collaborative filtering and hybrid filtering to recommend potentially intriguing advertisements to users based on information such as user’s reactive behavior to advertisement, user’s personal characteristics and the content of the advertisement itself.Collaborative filtering algorithm is one of the most widely used recommendation algorithms. However, because of the special attributes of advertisements and users, application of collaborative filtering algorithm to the advertising alliance, also faces the following problems:(1) The real time matching between advertisements and changing user interests; (2) The timeliness of users’similarity; (3)The definition and measurement of user similarity.In addition, how to use the existing advertising platform (the use of its advertising data and user data resources)to build an advertising alliance that uses personalized advertising recommendation technology as its core technology and how to integrate the data acquisition and advertisement recommendation into an organic synthesis based on the online-near line-offline three layer architecture,, are two major architectural issues that need to be considered in this thesis.This thesis first analyzes the existing platform function and structure, and puts forward the reformation requirements. Next, this thesis introduces the idea of timing update, reliability and score-based predication and improves the Weighted Slope One (WSO) algorithm based on the traditional collaborative filtering. Finally, the modified algorithm and original algorithm are compared by experiments on randomly samples taken from the existing system, and the results show that as time passes by, the modified algorithm has better performance. |
