Research On Advertising Recommendation Model Based On Exposure Prediction

In recent years,as the momentum of digital marketing has risen sharply,based on the rapid development of Internet technology,advertising has become one of the profit points of many platforms,and advertising space has gradually become an important means of traffic realization.How to use exposure to show advertisers the effect of ad placement has become one of the key research issues in the field of ad recommendation.Ad exposure is jointly determined by ad click rate,ad conversion rate and ad bidding,because ad exposure estimation is related to the interests of users,advertisers and platforms,and whether the ad recommendation system can operate stably in the long term after it is put into effect requires not only a reasonable bidding plan with reference to exposure but also a reasonable ad placement plan with reference to the stability of the system.Therefore,in this paper,we make a prediction of ad exposure and conduct an in-depth study on the stable operation of the ad recommendation system.For advertising exposure estimation model,this paper proposes a Graph_CIN based exposure estimation model.Firstly,the advertisement exposure log data are pre-processed,the global and local features are mined by Graph Embeddig method,the low-order and high-order features are fitted by CIN network,the important features are extracted by LightGBM model,and the extracted feature vectors are input to the MLP model to estimate the exposure of advertisements.For the stable operation of the recommendation system,this paper proposes the design of H_∞ controller based on the robust control algorithm and solves the control law by simulation.From the perspective of control theory,we focus on the modeling and robust H_∞ control dgorithm of a class of ad recommendation system,and give the specific design of robust H_∞ feedback control law in the framework of linear matrix inequality(LMI),which can better solve the problem of external interference during the operation of ad recommendation system and provide advertisers with the decision basis for ad placement.Based on the desensitized data provided by real social platforms,this paper firstly performs data pre-processing,secondly extracts features from business rules,statistical indicators and graph embedding,and inputs them into the estimation model,simulates and solves the proposed system controller according to the estimation results,and proposes ad placement suggestions according to the simulation results.The research results show that the ad recommendation model for exposure estimation proposed in this paper can provide a reference for solving the interference problems that occur during the online operation of the exposure estimation and ad recommendation system for advertisers in reality.