Research On Personalized Recommendation Algorithm Based On Deep Learning

The rapid development of Internet applications such as online entertainment and online shopping enriches people's entertainment life.However,users also need to spend a lot of time and energy searching for their interesting content in the mass information.Therefore,the recommendation algorithm came into being.With the increase of user interaction and the improvement of product information,the volume of data increases rapidly and the data sparsity increases dramatically.Traditional recommendation system can not fully extract the characteristics of users and items,and also can not fully explore the internal relationship between users and products,which leads to the unsatisfactory recommendation effect.Although the traditional recommendation algorithm to solve some problems of user information search,but along with the increase in user interaction,the improvement of the product information,data volume increases quickly and sparse data has increased dramatically,the traditional recommendation system unable to fully extract the characteristics of the users and items,at the same time also can't fully explored the intrinsic relationship between users and products,lead to recommend the effect is not ideal.Deep learning has a powerful learning ability,It can learn the essential features from massive sample data sources and has good portability.Therefore,more and more scholars apply it to the field of recommendation system.Convolutional neural network is one of the most widely used and effective neural networks in the field of deep learning.It can conduct in-depth mining and analysis of text information features,and then classify user features well.In recent years,it has also been gradually applied to the recommendation field.However,the existing research is not enough to deal with the information features at the information input end.For example,the non-user rating data such as user portrait,user behavior and product attributes are not classified and processed at the information input end.At the same time,the research on the feature fusion layer is insufficient.In view of the above problems,this paper makes an in-depth study on the feature extraction of the recommendation algorithm and proposes the following improvements:1.In view of the problem that the user and project information extraction is not sufficient in the recommendation system,the user and project auxiliary information are divided into shallow semantic information and deep semantic information,and a semantic classification processing neural network is proposed to fully extract the shallow semantic features and deep semantic features of user and project auxiliary information.2.In view of the feature fusion layer of the recommendation system based on deep learning,the deviation between users and projects is not taken into account,and the internal relationship between users and projects is not taken into account,so the following improvements are made in this paper:(1)In view of the problem that individual users have deviation in project rating,a new collaborative filtering algorithm based on convolutional neural network is proposed.By introducing user deviation,project deviation and global deviation into feature fusion layer,the user rating will not deviate seriously.The experimental results show that the improved algorithm has improved both the root mean square error(RMSE)and the mean absolute error(MAE)index.(2)In order to fully explore the complex relationship between users and projects,this paper proposes a feature fusion method based on the product form of quadratic polynomials.The method takes into account the relationship between users,the relationship between projects and the relationship between users and projects.The rigorous experimental results show that the algorithm is based on the root mean square error(RMSE)and the mean absolute error(MAE)means The standard is better than the current mainstream deep learning recommendation algorithm.