Research On Recommendation Model Based On Deep Neural Network

The traditional collaborative filtering recommendation algorithm uses the observed score value to predict the missing value.The method is simple and fast,but it ignores the possible interaction between variables in the data set and does not reflect the nonlinear relationship between the user and the project.There is a large error between the predicted score and the true score.Inappropriate estimates increase the noise in the data set and become worse as sparsity increases.It is also susceptible to cold start problems.In order to solve these problems,this paper introduces deep neural network to the recommendation algorithm,proposes two hybrid recommendation models,and uses a multilayer perceptron to learn the nonlinear relationship between users and projects to alleviate the data sparsity and cold start in the traditional recommendation algorithm.Problems,improve recommended performance.The first recommendation model is a deep neural network recommendation model DNCF that incorporates user and project information.Extracting attribute features of users and projects through neural networks and using the word bag model to process textual information can not distinguish the semantic features of the words in the text.A convolutional neural network is used to process the text auxiliary information,and the generation can effectively represent the text in depth.The potential features of semantics,while combining text features with their corresponding project features.For the traditional model to make the inner product of user and project features insufficient to predict the complex interaction between users and projects,a multi-layer perceptron is proposed to learn the interactions between users and projects,and to input user and project features The layer sensor predicts the score and provides the user with a recommendation based on the score.The second recommendation model m DAE combines user and project information with score data.Since the score data is sparse,a stack-type noise reduction automatic encoder is used to process the score data to obtain low-order dense features,and user and project attribute information is input into the stack noise reduction.Automatic Encoder,which obtains hidden features of users and projects,still uses convolutional neural networks to process text-assisted information,combines text features with their corresponding project features,and finally uses user and project features as input data for multi-layer perceptrons.Predict scores and make recommendations for users based on ratings.The experiment was designed according to the principle of the model.The two proposed algorithm models were tested offline on the Movielens dataset and the Amazon dataset.The experimental results showed that the two models proposed in this paper and the traditional commonly used CMF,PMF,SVD,and item Compared with the-based algorithm,the RMS error is reduced and the recall rate is improved.The two proposed models are compared and analyzed.Compared with the first algorithm,the RMSE on the four data sets is reduced,and Recall and MAP are improved.The combination of user-item rating data and user and project information can improve the accuracy of the recommendation,enhance the stability of the model,and have better recommendation performance.At the same time,it shows that the use of user and project auxiliary information as data supplement can effectively improve the data sparse situation,solve the cold start phenomenon,and further improve the recommendation effect.This paper mainly studies the application of the deep neural network in the recommendation model,studies the problem of data sparse and cold start,proposes two recommended models,and demonstrates the validity of the proposed model through experiments.