Research On Movie Recommendation System Based On Improved Neural Collaborative Filtering Model

Relying on the vigorous development of China's film market,the demand for personalized film recommendation system arises at the historic moment.It is more and more important to build a more accurate personalized film recommendation system.In order to overcome the limitations of traditional recommendation algorithm,this paper attempts to use neural network technology to carry out the research of recommendation system based on deep learning.In this paper,we use the data set movielens,which is published by GroupLens team of University of Minnesota(UMN),to test the recommendation algorithm,and improve the Neural Collaborative Filtering model from two perspectives to get two kinds of movie recommendation system models,mainly including the following work:(1)An improved B-NCF model combined with Bayesian Personalized Sorting algorithm is proposed.By constructing user movie partial order pair,satisfying the precondition of Bayesian Personalized Sorting algorithm,replacing GMF layer in the original NCF model with BPR layer by redesigning NCF neural network,NCF neural network can obtain the ability to learn the implicit sorting information,so that the Bayesian Personalized Sorting algorithm and NCF model can be cleverly combined.In this paper,the validity of the new model is verified by a large number of comparative tests.The introduction of BPR layer enhances the ability of mining information depth of recommendation model,and at the same time increases auxiliary information to affect the final recommendation score,so the final movie recommendation system meets the requirements of movie recommendation from depth and breadth.(2)A multi information embedding recommendation model M-NCF is proposed.Because the data storage formats of MovieLens-100 k and MovieLens-1m are different,different data processing codes have been written.In order to extract the effective digital features,a set of digital scheme for MovieLens data set is proposed,such as age and occupation,which can be directly mapped to the correspondingnumbers.The movie name is extracted from the text information through digital mapping and Text Convolution Network.Solve the problem of multi information fusion through this digital solution;three kinds of neural networks are used to process multi information,mining the text information and digital information that can be obtained from data set,and finally a multi information embedding recommendation model is obtained by connecting all layers of neural network information with full connection layer,which solves the traditional problem.It is proved that the learning ability of the new model is stronger and the accuracy is higher than that of the original model by designing a number of comparative experiments,and a variety of recommendation schemes are realized.