Research On Association Model Based On Deep Learning And Its Application In Recommendation Field

With the rapid development of the mobile Internet,it is more and more convenient to watch movies.From the end of the nineteenth century to the present,the types of movies have become more and more abundant,and the number of movies has also increased.Not only the traditional film industry will launch a large number of new movies every year,but also domestic media platforms such as Tencent,i Qiyi,and overseas media platforms such as Netflix and Apple.A large number of self-made works will be produced,and the overload of film and television information makes the choice of users more and more difficult.Applying the recommendation system to the field where there is a large amount of video can not only improve the user's viewing experience,but also promote the development of the recommendation system through user feedback data.In recent years,deep learning has made great breakthroughs in various fields,especially in the performance improvement of traditional recommendation algorithms,which has played a very obvious effect.The usual practice is to introduce auxiliary information other than scoring into the recommendation algorithm,which can more effectively use various data information and solve the problem of data sparseness.However,the use of auxiliary information is currently limited,and the deep-seated features of the data have not been deeply explored.Through in-depth analysis and research of recent deep learning algorithms applied in the recommendation system,an improved deep learning network movie recommendation algorithm MCDNets based on multi-layer perceptron(MLP)is proposed,which can effectively solve the problem of information overload.Different from the simple use of auxiliary information and the prediction method of vector multiplication by traditional hybrid recommendation methods,the algorithm uses convolutional neural networks to extract the context information of the project description document,and input the extracted user features and project features into the multi-layer perceptron to Predictive ratings improve the deep network recommendation model.After that,an improved convolutional collaborative filtering network UCCFNets was proposed.The MCDNets algorithm was used to extract the nonlinear shallow feature to prediction scores,and the improved matrix decomposition algorithm was used to extract the deep feature to prediction scores between users and items.Through the weighted association mode,the two are effectively combined to generate a new score.By conducting comparative experiments on multiple recommendation algorithms such as probability matrix factorization(PMF)and convolution matrix factorization(Conv MF),on three real data sets with different sparseness,the newly proposed algorithm not only has accuracy in score prediction The improvement also alleviates the cold start problem to a certain extent,and in the case of sparse data,the improvement of the model is greater.