Research On Secure Attribution Model Based On Data Fusion

Advertising attribution is tracing the advertising information,mining and measuring the value of different advertisements to users.it is an important basis for advertisers to make budget decision and evaluate the marketing value of advertising touches,which is of great significance to advertisers and advertising platforms.However,most of the existing advertising attribution solutions are based on the assumption that the attribution party can independently complete the attribution model training and attribution weight distribution based on all the information required for attribution,without considering non-unique data sources and the secure issues such as privacy data leakage in actual attribution.Moreover,in terms of attribution modeling,the existing attribution schemes lack comprehensive consideration of the interaction between ad touches in the ad-side access sequence,the influence of ad’s attribute information on the ad touches attribution weight distribution and user transformation,resulting in poor attribution performance.Based on these,in order to realize the secure ad attribution of multiple data sources under protecting privacy,this paper studies a secure ad attribution model based on data fusion.The main research works and innovations of this paper are as follows:(1)Aiming at the problem that the attribution data sources of advertising platform are scattered and difficult to aggregate,this paper proposes a data fusion attribution architecture based on homomorphic encryption-p-2PDF.The architecture uses Paillier-based homomorphic encryption technology to construct a joint attribution model between advertisers and advertising platforms,and safely and effectively integrates multi-party data for attribution to achieve ’ data is usable but not visible’.Experiments on real data sets show that this architecture can achieve two-party data fusion attribution modeling that protects privacy without losing the attribution accuracy of the model itself.(2)Aiming at the problem of insufficient mining of ad-side features by attribution model,this paper proposes an advertising attribution model based on bidirectional LSTM combined with attention mechanismABLSTM.The model combines bidirectional LSTM and attention mechanism to mine the interactions between channels in sequence features,and uses the embedding layer combined with the fully connection layer to process the feature information of advertising exposure attributes,so as to realize multi-factor advertising attribution.By comparing ABLSTM with other attribution models,the effectiveness of ABLSTM attribution model in predicting user conversion is verified.(3)This paper designs and implements a secure attribution system based on data fusion.It mainly includes data acquisition layer,data analysis layer and display layer.The acquisition layer mainly obtains multi-type advertising data and user conversion data.The analysis layer mainly runs a multi-data source attribution model.The presentation layer shows the results of the data analysis layer through interaction with the user ’s UI.The test results show that the system meets the needs of the advertising platform to complete the accurate attribution.