Research And Implementation Of Personalized Movie Recommendation System Based On Spark

The recommendation system is to help users find the information they need conveniently and quickly,and act as a bridge connecting users and information.In reality,since it is difficult for users to evaluate a large number of items,a dense scoring matrix cannot be formed,which is not conducive to the recommended calculation operation.When similarity needs to be calculated in neighborhood-based collaborative filtering,the recommendation effect may be greatly reduced due to the lack of common historical behavior information.On this basis,this paper uses a model-based collaborative filtering algorithm to mine the hidden potential information between users and items,which can effectively alleviate the problem of data sparsity.Despite this,there is still a cold start problem.When a new movie enters the system,it is difficult to push it to users who need it.In addition,the user's interests and hobbies will also change accordingly with the passage of time.For the cold start problem,when new movies enter the system,it is difficult to push them to actual users.For this reason,a collaborative filtering recommendation algorithm based on fusion clustering and matrix decomposition is proposed.By integrating the idea of clustering into the traditional matrix factorization recommendation algorithm to achieve the effect of alleviating this phenomenon,first decompose a large and sparse rating matrix into two smaller dense matrices to find the set of neighbors of the target item,And then use their corresponding attributes to fill in the new item attributes in a certain way,and then update the matrix data information to make effective recommendations for users.For the problem that users' interests and hobbies will change over time,a collaborative filtering recommendation algorithm based on time factors and item attributes is proposed.Regarding the characteristics that people's interests and preferences will change with the passage of time,this paper uses the forgetting function curve to simulate the process of human brain forgetting.In the traditional collaborative filtering recommendation algorithm,the user's rating time factor is incorporated into the similarity of items.In the calculation,it is combined with the item attribute similarity to obtain a comprehensive item similarity calculation method,which can more truly show the current state of interest and hobbies.This article builds a recommendation system for movies on the Spark platform,and also uses the Spark MLlib ecological library to implement it.This system will be designed from offline,online,and popular recommendations.The system in this paper can make full use of the user's implicit and explicit behavior information to provide users with recommendations that meet their own characteristics.Through corresponding experiments,it can be seen that the effectiveness of the algorithm improvement proposed in this paper and the feasibility of the recommended system designed can meet the needs of users.