Research On Hybrid Recommendation Algorithm Based On Deep Learning And Stacking Integration Strategy

The recommendation technology can be used without user's clear purpose,by analyzing the user's past purchase behavior,creating a user preference model and digging the long tail of the items.The more suitable items that users may be interested but difficult to be found can be recommend to the users.However,due to the rapid development of current deep learning and mobile Internet computing technology,the massive data stored on the Internet has increased exponentially.Although the data generated by users can help the recommendation technology to obtain more effective information and can greatly improve the recommendation effect of the recommendation algorithm,a large number of users arbitrarily created label data to the recommendation technology is faced with data sparse,scalability,cold start,accuracy and effective feature extraction difficulties and other problems,resulting in poor user experience and low recommendation accuracy.The content-based recommendation algorithm,the probability matrix decomposition and the user-based recommendation algorithm are analyzed.In traditional recommendation algorithms,user-item rating data becomes more and more sparse.The collaborative filtering method is the most widely used recommendation technology.It uses a simple inner product interaction model,only considers a single rating information,has a cold start problem,and cannot learn the complex nonlinear structural features between users and items.However,the content-based recommendation algorithm has the advantage of personalized,but it can only recommend items similar to the user's past browsing,and encounter difficulties in effective feature extraction.The characteristics of deep learning models can learn more complex structures,so it has become a new development trend to integrate deep learning technology with recommendation algorithms to solve the above problems.In response to this problem,a hybrid recommendation model is proposed based on deep learning and Stacking integration strategy.The Stacking integration strategy is used to first fused user-based,probability matrix decomposition,and content-based recommendation algorithms,generalize the output results of multiple models,so that the algorithms can complement each other and break through the bottleneck of a single model,so as to achieve better recommended performance.The fusion-based model learns the more abstract and deeper nonlinear interaction features by deep learning technology,alleviating the problem of data sparsity,and makes the model performance gain further.The algorithm is verified by using Python programming.Compared with the traditional recommendation algorithm user-based,probability matrix decomposition and the content-based on the MovieLens-100 k,Movie Lens-1m and Pinterest datasets,the results show that the proposed hybrid recommendation model has significantly improved the accuracy of rating prediction on the test dataset.