Research And Application Of Personalized Recommendation Based On Deep Learning

With the expansion of information technology,the issue of information overload is serious.Users need to spend a lot of time and energy to find information that meets the demand.It is difficult for information producers to rack their brains to provide recommended information for users,and the user experience is poor.How to quickly and accurately screen out the information that meets the needs of users from massive data is a challenging task.The rapid development of the recommendation system has made it possible to solve these problems.The recommendation algorithm often draws on the methods of machine learning and deep learning,and has achieved rapid development,and has also promoted the development of other fields.This thesis studies the recommendation algorithm and application,and combines the incremental frequent pattern excavating algorithm with the deep learning model WDL(Wide and Deep Learning)to provide automatic cross-feature learning.The structural learning between features is introduced into the WDL model,and a FWDL(Frequent Wide and Deep Learning)model is established.In order to achieve this goal,the main contents of this thesis are as follows:1.The development status of the recommendation algorithm is analyzed,and the application and advantages of deep learning in the recommendation field are explored.The WDL recommendation model is analyzed.2.The incremental frequent pattern excavating algorithm EFUFP is suggested based on support number,and the excavating of frequent features under large volume of data is completed,and it is applied to the recommendation system for frequent item mining and relation mining.3.Based on EFUFP algorithm and WDL model,FWDL algorithm is proposed to fuse frequent pattern mining with deep learning,and a hybrid recommendation model FWDL model is established.The model provides the learning of automated frequent cross-characteristics and relationships without the need for manual feature engineering.4.The performance of the EFUFP algorithm and the FWDL algorithm are verified separately,and comparative experiments are performed on multiple data sets.The results show that under the premise of ensuring correctness,the operating efficiency of the EFUFP algorithm is improved by 10.7%in the local environment and by 26.8%in the Spark platform.Compared with the original model,the FWDL algorithm improves the AUC by 1.16%,the accuracy by 1.22%,and the recall rate by 2.76%.5.Based on the above research,the Android-based medical health knowledge recommendation system was designed and implemented,and a health knowledge dissemination platform was built to test the accurate recommendation application ability of the FWDL algorithm.It can provide users with efficient personalized recommend-dation services,saving users' access to medical health related knowledge and search time.