A New Recommendation Of GAN Based On Time Series

With the development of science and technology,the popularization of scientific and technological products,as well as the arrival of the era of big data and other cross-era changes,the amount of data information that people can use is also increasing.How to quickly and accurately find the information they are really interested in in the chaos of information has become a vexing problem for every user.The search engine born for this purpose can filter information through the keywords entered by users,greatly speeding up the speed of information filtering by users and solving the problem of "information overload" to some extent.However,in the case of unable to accurately describe their own needs,search engines can not provide quality services.As a result,personalized recommendations that are more userfriendly have been born.Personalized recommendation refers to the analysis of project attributes,user behavior and other characteristic information through various recommendation algorithms,and then provide users with the items they really like through the recommendation system.In recent years,deep learning,as an emerging branch of machine learning,has become the mainstream of Internet big data and artificial intelligence.In the field of recommendation system,the fusion of recommendation system and deep learning model is also concerned.Compared with the traditional recommendation algorithm,the recommendation algorithm based on deep learning model integrates the advantages of deep learning.For example,automatic learning of abstract features of users or items,the ability to use multiple types of data as input,and the ability to extract more hidden features than traditional recommendation systems.All these can make the model better identify user preferences and improve the accuracy of recommendations.This paper is mainly based on deep learning and recommendation system.Through in-depth research on deep learning model,recommendation algorithm and recommendation algorithm based on deep learning,it proposes to apply the generation adversarial network integrated with time series to the recommendation system.It is mainly used for user-project implicit feature extraction and user-project score prediction.At the same time,according to the idea of cyclic neural network,this paper also proposes a personalized recommendation algorithm based on cyclic generation adversarial network.The main tasks are as follows:(1)for the research purpose of this paper,in-depth understanding of personalized recommendation algorithm and the current popular deep learning model,focusing on the model with significant effect on feature extraction,unsupervised learning and time feature transmission.At the same time,a lot of research has been done on collaborative filtering and matrix decomposition algorithms for effect comparison.It lays a solid theoretical foundation for this study.(2)put forward the index of comment quantity difference.Due to the traditional collaborative filtering algorithm is by cosine similarity method,calculate the deviation between the user evaluation vector as similarity.This method in the two user comment number is large,relatively the credibility of the similarity will be reduced.Therefore,the similarity of users with a large difference in the number of comments can be improved by using the difference index of the number of comments,which to some extent solves the similarity deviation caused by the difference in the number of comments.(3)the generative adversarial networks incorporated into the time series is applied to recommendation.This is one of the core of this paper.The convolution neural network is used to extract and generate the concept of confrontation network,which is integrated into the time series.In order to avoid the complex hidden feature extraction problem.At the same time,the integration of the generator and the time series can make the recommendation results not only unitary but also traceable.(4)through the project classification or clustering,substitution matrix in the method of the project,to a certain extent solved the recommended coverage is not high and the training time problem.The recommended coverage rate of generative adversarial networks is not high,which is mainly limited by the size of the matrix.Therefore,in the case of unable to expand the size of the matrix,replace training matrix data,in order to store the results of different training model.(5)put forward the cyclic generative adversarial networks.Combining the idea of circulating neural network and generating adversarial networks,the time information of data can be transmitted directly in the model.