Study Of Diversity And Dynamic Drift Problem In Collaborative Filtering Recommender System

With the popularization of computers and the rapid development of Internet technologies,the information on the Internet grows fast.In that case,information overload becomes an urgent problem.The traditional search engine technology can't meet users'requirements because of its passive way of offering service,ignoring individuation and so on.As a result,recommender system is created.At present,collaborative filtering recommendation achieves lots of notices and researches as one of the most successful data mining technologies.It can be found that these researches mainly focus on accuracy but not diversity and dynamic drift problems of collaborative filtering recommender system by searching literature,so this article did some researches on these two problems,the main research work is as follows:(1)The article introduced the original intention and definition of recommender system.Then it summarized the mainstream recommendation algorithms,especially introduced the collaborative filtering recommendation algorithm,including the algorithm's thought,classification and steps of recommendation.At the same time,the article showed the frequently-used evaluation index,including accuracy evaluation index and diversity evaluation index.(2)For the diversity problem of collaborative filtering recommender system,this article designed the collaborative filtering recommendation algorithm based on item categories.This algorithm was based on item-based collaborative filtering recommendation algorithm,and took item categories information into consideration.It defined the function called item categories' contribution and used it to improve the formula of predicting ratings.The theory of the algorithm was to decrease the ratings of those items which had the same item categories with the item waiting to be rated,but increased others which might have some same item categories with it to improve the diversity of recommendation.At last,the results of the experiments showed this algorithm improved the diversity of recommendation.(3)For the dynamic drift problem of collaborative filtering recommender system,existed improvements mainly focus on the changing of users' interests,goods' life period and time hotspot's effects.They ignore the effects of goods with periodic-bought characteristics on recommendation.For this phenomenon,this article designed the collaborative filtering recommendation algorithm based on time weighting periodically.The algorithm was based on time weighting and took the effects of time period factor into account creatively.It defined the function called periodic contribution and used it to improve the formula of predicting ratings.The theory of the algorithm was to strengthen the impacts of ratings given in the active time on recommendation to achieve a higher accuracy.At last,the results of the experiments proved the improved algorithm had a higher accuracy.This paper did some researches on diversity and dynamic drift problems of recommender system.Then it introduced the collaborative filtering recommendation algorithm based on item categories and the collaborative filtering recommendation algorithm based on time weighting periodically.The former was unlike general improved algorithms which started from the results to improve.It started from the source to find the method of improvement.The improved algorithm was suitably used in sceneries which had high requirements for the diversity of recommendation.The latter took periodic actions into consideration for the first time in addition to changes of users' interests.The improved algorithm had an obvious advantage and a significant prospect in sceneries with periodic-bought characteristics.Finally,experiments proved the two algorithms'availability.