Mobile Application Recommendation Based On LightGCN And Multimodal Feature Fusion

With the rapid development of 5g,industrial Internet,and mobile network,mobile devices have become an indispensable medium in people’s daily life,and the number of mobile applications is growing explosively.In the face of many mobile applications with rich information,it is difficult for users to find suitable mobile applications.Therefore,it is necessary to recommend personalized mobile applications to users.Recommendation system plays a vital role in helping users get rid of the overload of mobile applications and automatically recommending appropriate mobile applications for users.Existing studies have proved that effective mobile application classification methods can improve the accuracy of recommendation systems,and the application of machine learning and deep neural network technology to mobile application classification and recommendation has become a research hotspot.However,the existing methods do not make full use of the multi-modal features of the mobile application contains information and the fine-grained interaction between high-order and low-order features,and the convolution operation in the deep neural network technology cannot well fit the distinctive spatial structure features such as the atlas abstracted by the recommendation system.Although the existing methods have achieved remarkable results in mobile application recommendation,the accuracy of recommendation needs to be further improved.To solve the above problems,this paper proposes the following two mobile application recommendation methods:(1)Using the interaction information between users and mobile applications,this paper proposes a mobile application recommendation method based on lightweight graph convolution network.Firstly,the interaction between users and mobile applications is modeled by a bipartite graph.Then,the features on the graph are smoothed by using the lightweight graph convolution network,the high-order connection between the user and the mobile application is extracted,and the feature representation of the user and the mobile application is generated by using three convolution layers.Finally,the inner product predicts users’ preferences for different mobile applications and completes the recommendation task.Using Kaggle’s actual data set,Shopify-app-store,several comparative experiments are carried out.The experimental results show that this method is superior to other methods in precision,recall and ndcg.(2)Using the multi-modal information of mobile applications,this paper proposes a mobile application recommendation method based on multi-modal feature fusion.Firstly,mobile applications’ image and description features are extracted using the inner volume residual network and the pre-training language representation model,respectively.Secondly,the attention mechanism in transformer model is used to fuse the image and description features of mobile applications.Then,the softmax classifier is used to classify the mobile applications according to the fused features.Finally,based on the classification results of mobile applications,the bilinear feature interaction model is used to extract the high-order and low-order embedded features of mobile applications,and the Hadamard product and inner product are used to realize finegrained high-order and low-order feature interaction,update the mobile application representation and complete the recommendation task.Several groups of comparative experiments are carried out on the real data set 365 kIOS apps dataset of Kaggle.The experimental results show that this method is superior to other methods in macro F1,Accuracy,AUC and logloss.