Research On Recommendation System And Apply It Into Mobile E-commerce

With the computerization of society, the production and ability to collect data significantly enhanced. Mass information brings us many opportunities, but it also causes many thorny problems, the biggest one is message filtering. Personalized recommendation system is a retrieval tool that more smarter than search engine: users do not need to provide keywords, system can automatically record and analyze user’s historical behavior data, and then present the information to user who will be most interested in that.After the recommendation system appeared, it has been widely used in e-commerce, but it also faces many challenges, such as many algorithms do not model time-information into the recommendation systems, thus cannot capture many important time-related patterns, this is so called static recommendation. To solve this problem, this thesis firstly intensively studies all kinds of recommendation algorithms. Then, probe the influence that time-information impact on recommendation results. Third, fuses time information into Collaborative Filtering, and puts forward two algorithms and gives the specific flow. Last, founds a recommendation model that fuses the mobile device characteristics, and applies this model into an online book sale system. The main works of this thesis are following:(1) Intensively studies the current main recommendation algorithms, then analyses the mechanism, advantages and disadvantages of every algorithm. Furthermore, gives some simple examples to describe each algorithm. Last, puts emphases on Collaborative Filtering and describes its recommendation process;(2) By analyzing time regular pattern in real life as well as time phenomenon that hides in Netflix dataset, this thesis demonstrates the fact that time-context information is very important to recommendation system;(3) By modeling time information into Collaborative Filtering algorithm, this thesis proposes two algorithms, one is collaborative filtering algorithm based on user impact factor(called IOE-User CF), another is collaborative filtering algorithm based on item coupling and popularity(called PC-Item CF), then uses Netflix dataset to test the two algorithms via some indexes, such as MAE, precision, recall and F1. Experiments shows that the Collaborative Filtering algorithm which mixed with time information has better quality than old algorithms;(4) Finally, this thesis establishes a dynamic recommendation model based on above idea, and fuses the advantage, that the mobile devices can easily get user contacts, into this model.(5) To demonstrate the above model has practical value, this thesis designs a simple online book sale system, and applies this model into this system.