Research Of Recommendation Algorithms Based On Grnn

With the advent of the era of big data,information is exploding.How to obtain value from such a huge amount of information has become a major challenge in the current Internet era.Personalized recommendation systems have emerged as the times require.It is mainly based on the user's information and historical purchase records and other data to recommend,with the development of e-commerce gradually popular.Although personalized recommendation technology is becoming more and more mature,there are still many bottlenecks,such as sparseness of data and cold start problems.This paper starts with solving the sparseness problem of data and uses GRNN to populate the user scoring matrix to improve the performance of the recommendation system.This article mainly proposes two methods to improve the accuracy of the recommendation system:1)The problem of matrix sparsity exists in the traditional collaborative filtering recommendation algorithm,which will cause the recommendation system to look for user neighbor nodes.Because the number of items that two users jointly score is too small,the similarity calculation is inaccurate.In this paper,GRNN is used to pre-populate the user's rating matrix,and the similarity between users is calculated by using the filled matrix.Because the number of users of the matrix after the padding is increased,the similarity will be more accurate.Compared with traditional neural networks,GRNN has a faster budget because it only has a four-layer network structure.In addition,it does not need to continuously train the weights and offsets of neural networks.It only needs to train a smoothing factor.In this paper,genetic algorithm is used to find the smoothing factor of GRNN.Genetic algorithm is a computational model that simulates the natural selection and genetic mechanism of Darwinian biological evolution.It is a method to search for the optimal solution by simulating the natural evolutionary process.After the GRNN is used to fill in the user rating matrix,the number of items that the user evaluates together increases,and the similarity between the users is also more accurately calculated.When the recommendation system looks for similar users,the user's neighbors are more reliable.2)The traditional method generally uses the distance between two users to measure similarity.This paper uses an improved information entropy algorithm to calculate the similarity,but the information entropy only considers the probability of scoring differences,but does not consider the size of the score gap.The impact,so this paper adds a scoring difference to improve the similarity.In addition,if the number of items jointly evaluated by two users is greater,then the two users are more similar,so this paper adds a Jaccard distance to correct the similarity.