Research And Application Of Mobile Advertisement Recommendation Technology Based On User Profile

Nowadays,more and more people has their own mobile devices,mobile advertising market share is also growing.Compared with the traditional Internet advertising,mobile media itself has the characteristics of mobile,fragmented,personalized,shared and participatory,mobile advertising must be placed in the direction of accurate and personalized development.Only by changing the traditional advertising extensive mode of delivery,for different users of different interests of personalized advertising,in order to convert ads into consumer behavior,so that advertisers and customers have a good business return.The existing personalized advertising recommendations are mostly based on the content of the recommendation,by extracting the user's current page keywords to match the ads,and did not consider the user's own interest.In this thesis,I use the RBF neural network algorithm to optimize the scoring matrix for the common problem of sparseness of data,and design a hybrid recommendation algorithm combined with user image modeling,which is based on the user's historical behavior data.Improve the recommendation accuracy,to achieve personalized mobile advertising recommendations.The main work of this thesis is:1.Analyze the user data collected by the system and perform accurate portraits.In the context of HERE Australia community activities platform project design a set of user portrait tag system,at the same time for the construction of user portrait modeling used in the VSM algorithm was improved,and discussed the portrait model update method.2.In-depth study of RBF neural network structure to optimize the scoring matrix,reduce data sparsity,and then put forward a combination of RBF neural network and user portrait modeling hybrid advertising recommendation algorithm.The core is to use RBF neural network to approximate the characteristics of nonlinear function with arbitrary precision,to predict the vacancy term of scoring matrix,and then use the Pearson similarity calculation method to get the user's largest neighbor set.The results of the preliminary recommendation are combined with the user model vector to calculate the similarity between them,and then improve the accuracy,and finally get a satisfactory recommendation.3.Based on the user portrait modeling method and the optimized hybrid recommendation algorithm proposed in this thesis,the mobile advertising recommendation system based on user portrait is designed and implemented under the background of the event community platform.The system is the main component of the platform of the Australian community,which is the main profit model of the Here platform.It also provides mobile micro-mail and personalized advertising recommendations for the platform,as well as integrated back-office configuration management of advertising.