Research On CTR Prediction Model Based On Multi-modal Signal

Click-through rate(CTR)prediction refers to the probability that users click to browse the information after the information is pushed to users,while in this paper,CTR prediction refers to search and e-commerce advertising.The research shows that the probability of users' clicks has a great correlation with the placement of information.In order to maximize the revenue,it is necessary to place the information with a large click rate near the top of the page of users' browsing.However,accurate CTR prediction can effectively utilize advertising space resources and bring considerable economic benefits to the advertising platform.At the same time,in the era of information overload,for users,CTR prediction can also help users save time in searching for information by means of accurate recommendation.Traditional CTR prediction model mainly relies on category features and numerical features to represent an advertisement content,which can no longer adapt to the current graphic information.Text description and image quality,content,have an impact on the click rate.In order to better integrate these multi-modal signals into our CTR model,the following studies are carried out in this paper.Secondly,aiming at the defects in model design and insufficient utilization of image features in the process of introducing image signal features into CTR prediction,a new CTR model based on deep neural network is designed in this paper.Finally,experiments on the data set prove the validity of our model.