Research On Mobile Internet Advertising Recommendation Algorithms

Due to the rapid development of Internet and mobile phone industry,the number of mobile phone netizens has increased these years.As of June 2018,in China,the mobile Internet users reached 788 million.The proportion of mobile phone netizens has reached 98.3%.The huge number of mobile Internet users brings potential business opportunities to mobile Internet advertising.It has gradually become an important advertising channel for online advertising.Facing such a huge number of mobile Internet users,It attracts many advertiser to enter the mobile Internet market and produces a large number of online advertisements.However,at present,many online advertisements are randomly placed,and the efficiency of it is very low.Meanwhile,the randomly placed advertisements interfere with users' normal life and online experience,arousing user's disgust.How to improve the accuracy of online advertising and reduce the cost is becoming more important than ever.Therefore,this paper proposes a mobile Internet advertising recommendation method.The main contents of this paper are as follows:1.Firstly,analyze user text so the user text similarity matrix is obtained.Then,according to the user matrix to obtain the advertising behaviors of users who are similar to the user text,and combine with the time penalty factor,generate the "user-advertisement list" matrix,and the user advertisement recommendation list is obtained by collaborative filtering.And then find out the users in the same area from the text similarity according to the geographical location of users,and then conduct collaborative filtering recommendation according to the "useradvertisement" matrix to get the new advertising recommendation list.Finally,by using linear weighting method to deal with these two advertising recommendation lists,the final advertising recommendation list based on collaborative filtering is obtained.2.Using the basic information of users(age?gender?region,etc)to construct Neural Network Model for Obtaining User's Characteristic Vector;Using the basic information of advertisement(advertisement name,advertisement type)to construct Neural Network Model for Obtaining Feature Vector of Recommended Objects.In this way,the neural network model of advertising recommendation can be realized.The user eigenvector and the feature matrix of the recommended object are used to calculate the score of the target to be recommended object and the obtained score is used for regression,to reduce the loss and to modify the weight and other parameters.Finally,the trained user eigenvector and advertising recommendation matrix are calculated to obtain the user's recommendation value for the advertisement.The first top-n advertisements of the recommendation value generate the user's advertisement recommendation list,and realize the recommendation of advertisement to users.3.To further improve the effect of recommendation results,the model fusion method is used in this paper.Advertising recommendation list based on deep learning and advertising recommendation list based on collaborative filtering are weighted linearly to get final advertising recommendation list of this paper.Finally,the first TOP-N advertisements in the list is selected,and these advertisements are recommended to users.Finally,the validity of the proposed algorithm is verified by comparing the model fusion before and after the model fusion.At the same time,the experimental comparison and verification of precision and recall rate are carried out with different recommendation methods,which proves the recommendation effect of the recommended method in this paper.