Research On Hybrid Depth Recommendation Algorithm Integrating Time Weight

Recommendation algorithm is to recommend products that users may like by using a series of user behaviors and relevant algorithms.Tiktok has gradually penetrated into the trend of social life.The familiar applications of jitter,Taobao,bean paste,Net Ease cloud,and the United States mission are based on recommendation algorithm.In fact,Traditional recommendation algorithms are difficult to meet the increasing data and user needs,There exist some problems of data sparsity,cold start,interest offset,feature engineering,which lead to poor recommendation effect and great limitations.This dissertation mainly proposes solutions to the problems of data sparsity and interest offset in traditional algorithms,The advantage of deep learning technology is used to alleviate the problem of data sparsity.At the same time,combined with time factors to capture users’ preferences and solve the problem of users’ interest deviation.The main contents of this dissertation are as follows:1.Aiming at the problem of data sparsity in traditional recommendation algorithms,and fully mining the characteristics of user items and ratings,a hybrid depth recommendation algorithm HSAEM based on semi-autoencoder and multi-layer perceptron is proposed.Due to the sparse user scoring data,the algorithm first obtains the deep-seated characteristics of users and projects through semi-autoencoder and fusing relevant auxiliary information.For the simple inner product interaction method used in collaborative filtering method,it can not handle the complex nonlinear structure characteristics with users and objects.Then,the algorithm uses multi-layer perceptron to nonlinear fuse the extracted deep-seated features to complete the final score prediction.HSAEM algorithm effectively combines the semi-autoencoder in deep learning with the multi-layer perceptron model.Experimental results show that the algorithm performs well in experimental data,which fully proves that the algorithm can alleviate the problem of data sparsity.2.In HSAEM recommendation algorithm,users,items and scores are borrowed to capture users’ preferences for items,and the default user offset is unchanged.Aiming at the migration of users’ interests and preferences in the algorithm,Reset the user preference matrix by setting a time weight function that represents the degree of decay of the user’s interest in the item and incorporating it into similarity and scoring predictions.The final recommendation is then made by fusing the constructed user preference matrix model with a hybrid model based on a semi-automatic encoder and a multilayer perceptron.The hybrid depth recommendation algorithm of fusion time weight can be used to capture changes in user interest and alleviate data sparsity problems by using the HSAEM hybrid model.Experiments show that the algorithm with time is better than the algorithm without time,indicating that time factor can improve the quality of recommendation.Figure [27] Table [6] Reference [66]...