A Novel Collaborative Filtering Recommendation Algorithm

Nowadays the application of e-commerce recommendation system is more widely,recommendation algorithm has also been extensively studied as the core of the recommendationsystem,collaborative filtering algorithm is one of the most successful recommendationalgorithm,but there are three problems that data sparsity,cold start and scalability.This paper is toimprove the existing problems in the traditional collaborative filtering algorithm.The main work ofthis paper is as follows:(1)A collaborative filtering algorithm that comprehensive user and item ratings and features hasbeen Proposed.This algorithm combines the user ratings similarity and the user features similarity toselecte user nearest neighbors,then compute the user prediction ratings,at the same time,it combinesthe item ratings similarity and the item features similarity to selecte item nearest neighbors thencompute the item prediction ratings.Then the user prediction ratings and the item prediction ratingsare combined to produce the final recommendation.This a lgorithm not only solves the cold startproblem,but also alleviates the data sparsity.(2)Using genetic algorithms to select the optimal combination of parameters has been putforward.In the improved algorithm,we need to select the appropriate thresholds or weights when weselect the user nearest neighbors and the item nearest neighbors.Genetic algorithm can get the bestcombination of parameters in the recommendation algorithm as one of the methods that solving thecombinatorial optimization problems,thus to improve the accuracy of the recommendation result.(3)Using genetic algorithms to solve the combination of the user features problem.How toextract the features that influence the current recommended scene is the key issue of therecommendation algorithm.We can extract the useful features with genetic algorithm.(4)This paper is simulated with the Movielens datasets,the experimental result shows that theimproved recommendation algorithm in this paper is more accurate than the traditionalcollaborative filtering algorithm,this algorithm can still get a better recommendation effect when thedatas of the user rating are extremely sparse.