Research On Recommendation Algorithm Based On Deep Learning

In recent years,with the development of information technology,a lot of data has been produced in the network.The recommendation system automatically provides users with the information they need by analyzing the data related to users.This can alleviate the problem of information overload caused by massive data.However,with the continuous growth of data volume,it is difficult to further improve the performance of traditional recommendation system.Deep learning technology has the ability to process heterogeneous,multimodal and other feature data,and can automatically learn the deep features of data.Therefore,in the background of the traditional recommendation algorithm can not get a breakthrough for a long time,the research of deep learning recommendation system has gradually attracted more and more attention.This has become a research hotspot in the field of recommendation system.This paper focuses on the prediction of scores and Top-K recommendation in the field of deep learning recommendation.In order to improve the performance of recommendation,feature extraction,hybrid recommendation,attention mechanism and interest drift are studied.The specific research contents are as follows:First of all,through the analysis,user rating is not only affected by user interest,but also by user rating tendency and project quality.Based on LFM algorithm,an improved LF-TaQ algorithm is proposed.This algorithm can not only extract the hidden features of users and projects,but also extract the features of users' rating tendency and project quality.The algorithm processes the data from the whole point of view and extracts the low dimensional features of users and projects.This algorithm effectively makes up for the defects of content-based deep learning recommendation algorithm,and lays a foundation for the research of deep learning hybrid recommendation algorithm model.Then,new features are introduced into the content-based MLP recommendation model,and a hybrid recommendation algorithm model based on feature combination is proposed.The model trains the user project data and interactive data separately,which has good stability and interpretability.Through the feature extraction of different data,we can effectively overcome the influence of data sparsity and improve the accuracy of model prediction.Finally,in order to accurately represent the current interest characteristics of users,we propose a recommendation algorithm which integrates attention mechanismand interest drift.To solve the problem of interest drift,a data preprocessing method is proposed to eliminate the problem that the new data can not effectively modify the model when interest drift occurs.Then,on the basis of the attention mechanism model,the interest drift unit is added to optimize the feature representation of user interest from different perspectives and improve the performance of the model.In this paper,there are 31 figures,13 tables and 80 references.