Research On Recommendation Model Based On Deep Learning

With the rapid development of big data,cloud computing,artificial intelligence and other new technologies,Internet applications emerge in endlessly,resulting in exponential growth of network data,and large-scale data brings new opportunities and crises for the development of various fields.On the one hand,massive data information brings more valuable information to the recommendation system.On the other hand,for the traditional collaborative filtering recommendation system,it has been difficult to meet the user's recommendation needs.Sparse problem,cold start problem and interpretability problem are all problems that need to be solved in recommendation system.Therefore,we need new technology to replace the traditional technology in order to effectively solve the above problems.Traditional collaborative filtering recommendation can't make a reasonable explanation for the recommendation results because of the single score data,and then extract effective implicit features from a certain kind of auxiliary information to assist recommendation.Although it can improve the accuracy of single domain recommendation to a certain extent,to deeply mine the complex relationship betweenusers and projects,we need to study the integration of multi-source data into the recommendation system,and research The new technology is used to extract the comprehensive features from the multi-source data.At the same time,the interaction between the extracted user features and project features is studied to calculate the prediction score.Based on the above background,this paper proposes a deep learning model DCNNs for recommendation system,which integrates multi-source data.(1)Multi source data is integrated into the recommendation system based on deep learning.In order to make the recommendation results not limited by single source data,solve the problem of sparsity and cold start in the recommendation system,and study how to obtain the data such as tags and comments related to users and items to be recommended from multi-source data,this paper proposes a parallel depth network model DCNNs,which combines the characteristics of deep and breadth structure,integrates multi-source data in breadth,and uses depth learning model in depth Learning implicit characteristics.(2)Factor decomposition machine is introduced to realize the interaction between user features and project features.Factor decomposition machine can combine user and project features,and it is better to synthesize the hidden features obtained from multi-source data.By using factor decomposition machine,users can calculate the comprehensive prediction score of a project,and then improve the interpretability of recommendation results.Through the experimental verification of the above research,the results show that DCNNs model is superior to the selected baseline model in RMSE and Mae indexes,and can solve the inherent problems of the traditional recommendation system well.