Research Of Personalized Recommendation Based On Time Context And Attributes

Personalized recommender system plays a role in e-commerce, which is like a shopping purchasing guide in traditional shopping mode. It can not only save labor costs greatly, but also not be limited in time, location, or other factors. Using a personalized recommender system will improve the service quality of e-commerce website, increase user stickiness and will be conducive to accept more consumers. When a user has the behavior of browsing or buying something, his behavior data will be retained. Personalized recommender system uses these data to profile users, mine their interest and provides personalized buying advice to them. The quality of recommendation algorithm which determines whether the recommender system can accurately meet the needs of users is the core of recent studies. As one of the earliest and most successful technology, collaborative filtering recommendation algorithm has been widely used. But traditional collaborative filtering recommendation algorithms only use the user-project history rating information without considering other dimensions of data leading to affect the accuracy.This paper focuses on how to use time context and attributes of users or projects to improve the quality of recommendation system on the basis of summarizing previous researches. The main contents include the following aspects:① This paper describes the concept and composition structure of recommender system thoroughly, compares several existing recommendation algorithm and describes how they are implemented and their scope. The principle of collaborative filtering recommendation algorithm is the focus of this paper.② A new time context weighting function is proposed. Traditional time weighting function treats each history data record in isolation. This paper uses the memory activation theory of cognitive psychology and the principle of cognitive unit activation, energy spread in spreading activation model to solve this problem. A calculation formula has been proposed to measure how much influence of a project generated on the time weighting of target project. Combining with multi-stage forgetting curve, the time weighting function this paper presents is more in line with the changes in user interest③ A new personalized recommendation algorithm based on time context and attributes is proposed. Traditional collaborative filtering recommendation algorithms only use the user-project rating data and have cold-start problem. This paper uses the attributes of users and projects to calculate similarity to solve this problem and combines the time weighting function based on memory activation theory with every rate prediction formula to increase the recommendation accuracy. A new personalized recommendation algorithm based on time context and attributes is proposed at last.Experimental results show the algorithm this paper proposed has higher recommendation accuracy than traditional collaborative filtering algorithms which were based on item and forgetting curve.