Research On Recommender System Based On Neural Network

With the continuous expansion of the scale of e-commerce,the number and types of goods are growing rapidly,which takes a lot of time for customers to find the goods they want to buy.The customers,which are faced with a lot of goods,will feel confused,which is the problem of information overload.The recommender system is proposed to solve this problem,which is built on the basis of mass data mining to help e-commerce websites to provide customers with complete personalized decision support and information services.In recent years,the research of deep learning has been continuously deepened,which provides a new research idea for the research direction of recommender systems.The recommender system algorithm based on neural network has great advantages in accuracy,and can provide people with personal electronic commodity decision support and information service more perfectly,which is of great significance to e-commerce platforms and costumers.Based on massive data,the recommender systems not only require high accuracy,but also require very high efficiency,so as to meet people's needs.However,due to the high dimension and high sparsity of e-commerce information,the general recommender system is difficult to deal with that,and the computational efficiency and algorithm accuracy are poor.In view of the above problems,this thesis focuses on the neural network recommender system based on high-dimensional and sparse(Hi DS)data to improve the computational efficiency and accuracy of the algorithm.The main contributions of this work include the following:(1)We propose a neural network recommender model based on CPU.Due to constraints,most research is still based on CPU calculations.However,the support of CPU operation for Hi DS data is poor,especially for neural network model.To solve this problem,this thesis proposes a fast recommender system based on Auto Encoder neural network model.For the Hi DS matrix,this algorithm does not fill in the missing data,and only focuses on the effective data,so as to ensure the computational efficiency and accuracy of the algorithm.In addition,a parallel framework based on Hog Wild is proposed to improve the computational efficiency of the algorithm.Finally,the model is extended to the deep neural network,and still achieves good efficiency.The experiments show that the algorithm has a great improvement in the computational efficiency while maintaining high accuracy.(2)We propose a neural network recommender model based on GPU.GPU computing has natural advantages for large-scale data and has become the most mainstream application platform of neural network model.However,GPU has poor support for Hi DS data,so it is difficult to compute Hi DS data directly.Therefore,this thesis proposes a neural network model based on GPU-based Auto Encoder.This algorithm optimizes the underlying computation and realizes CUDA-based GPU Hi DS matrix computation,to improve the computational efficiency of the algorithm.The experimental results show that the algorithm is effective in Hi DS data and improves the computational efficiency.