Research And Design Of Mobile Application Reco Mmendation System Based On Deep Learning

The mobile phone application recommendation systems can recommend for users from massive applications by their personalized usage,which effectively improve the user's experience and help applications occupying the market quickly.However,the application recommendation systems mainly rely on application management software.Due to the large number of application management software,it's difficult for a single software to fully capture the preference of all users.It also can't accurately reflect users' preference by updating or deleting the applications.And the scenario of such recommendation system is limited to the management software which is inefficiency.In addition,the traditional recommendation algorithm has the problems of cold start,data sparsity,difficulty in utilizing auxiliary information of users and items directly and difficulty in capturing intrinsic complex relationship between features,which become a bottleneck restricting the performance of the recommendation system.Therefore,this thesis constructs a model with the popular deep learning technology based on user and item interactive information and auxiliary information for explicit and implicit feedback recommendation,which can obtain the user's preference to items.Since mobile cellular data can reflect the user's real preference,based on the recommendation model proposed in this thesis,we design a mobile application recommendation system which can deploy and manage the model on multiple platforms.The main contributions of this thesis are summarized as follows:1.For the problems of the traditional recommendation system,we propose an end-to-end recommendation model based on deep learning without additional feature engineering.This model consists two parts,which are used to capture low-order linear relationship and high order non-linear relationship between features.And we construct the expressions of the user and item by user-item rating matrix,which improves the generalization ability of the model.We build the recommendation model with the TensorFlow framework,and verify the performance of the model by explicit feedback on the Movielens-1M dataset.The experimental results on this dataset show that our model has good performance and the RMSE reaches 0.8367,which proves the effectiveness of the model.2.In view of the shortcomings in data and application scenarios of the current mobile application recommendation system,we use the mobile cellular data to build an application recommendation model.We firstly analyze the usage of mobile applications,and filter a certain amount of data with manual rules.Based on the proposed deep learning model,we construct an implicit feedback recommendation model:we adjust the input,output,neural networks structure,loss function,evaluations of the model based on actual data.We build the application recommendation model with the TensorFlow framework.Under the real data,our model reaches 0.9278 in HR@10 and 0.7897 in NDCG@10.The experiments show that our model can be used for both implicit and explicit feedback recommendation system and verify the feasibility of the model in application recommendation system.3.Based on the model we proposed in chapter 4,we construct model iteration module and application recommender module which can use the latest data to iteratively update the model and select valuable users to calculate the application preference.Then we build a mobile application recommendation system based on B/S architecture which can deploy and manage the application recommendation model on multiple platforms,providing reliable data for subsequent precision marketing.