The Recommendation Algorithm Based On User Behavior Trajectory

Recommended system has developed rapidly in these years, but mostrecommendation systems use data of strong relationship like purchasing or rating. Thedata mining based on the weak relationship based on the trajectory of user behavior,especially based on user browsing behavior has developed relatively slow. However,this part of data occupies more than80%of all the amount of information, which have ahuge potential to be mined. How to make more effective use of this part of data hasbecome an urgent problem.Existing research shows that there is a strong relationship between users’ browsingtrack and their purchasing behavior. And we can mine users’ interest from theirbrowsing track, the time spent of browsing and the time of user drags the scroll bar ofthe page. The main works of this thesis are as follows:1. This thesis regards the user purchasing behavior as the accumulation of userbrowsing. According to the above point of view, this thesis puts forward a conceptsimilar to the pheromone in the ant colony algorithm which is called pheromones ofproduct. According the pheromones of product, we can connect users’ browsingbehavior and their purchasing behavior to make effective recommend.2. Through experiments, we have compared the performance of the algorithmunder different parameter value and found a best method for determining the parametervalue. Meanwhile, we have compared the effect between the algorithm in this thesis andthe current popular recommendation algorithm. Through the experiments, we havefound that the algorithm in this thesis is advantage to handle the problem of cold startand sparse data.3. Finally, the algorithm has been implemented distributed and incremental, sothat the algorithm’ scalability and real-time has greatly improved. Not only that, thoughthe volatile pheromone, the algorithm is able to capture users’ new behavior andgradually eliminate users’ old behavior.