Research On Hit Rate Prediction Of User Interest Point Modeling

With the development of the times,the amount of data of all kinds of information in the Internet is increasing,and the traditional Click-through rate prediction methods can no longer meet the needs of the current era.It has become a trend to apply the deep learning model to solve the problem of Click-through rate(CTR)prediction.The main work of this article is as follows:To improve the expression ability of the Deep FM model,a Deep Learning Model of Item Feature Extraction Based on Attention Mechanism and Residual Network(DIAR)is proposed.The DIAR model continues the design philosophy of the Deep FM model and can mine data information from multiple levels.The DIAR model is divided into three modules: shallow feature extraction module,deep feature extraction module and Click-through rate estimation module.The shallow feature extraction module is implemented by a field perception factor decomposition machine model with domain features,which is used to extract the most essential information from the data.The deep feature extraction module is implemented by using the Convolutional neural network model.At the same time,the module also uses the residual module design idea,attention mechanism,and standard hidden layer output to improve the model’s ability to mine the deep features of data.The Click-through rate estimation module in the DIAR model is responsible for combining the shallow and deep features of the data,and outputting the estimated Click-through rate.The DIAR model is implemented under the Python deep learning framework.Compared to the Deep FM model,the GAUC value of DIAR on the Avazu dataset increased by0.3% and the Log Loss value decreased by 1.0%;The GAUC value on the Criteo dataset increased by 0.8% and the Logloss value decreased by 0.7%.In order to improve the expression ability of the DIAR model,a prediction model of deep learning click through rate for modeling user interest points(PDU)is proposed.The PDU model is divided into three modules: user feature extraction module,feature extraction module of items to be recommended,and Click-through rate estimation module.The user feature extraction module first analyzes the user’s behavior sequence and extracts feature vectors containing user interest information from the user’s behavior sequence;Then,the feature vector containing user interest information is combined with the feature vector containing user basic information to complete the primary modeling of user information;Finally,the PDU model achieves further information extraction through the DIAR model,completes user information modeling,and obtains user feature vectors.The feature extraction module for recommended items is directly implemented using the DIAR model,which is used to obtain the feature vectors of recommended items.The Click-through rate prediction module in the PDU model will analyze the user feature vector and the feature vector of items to be recommended,and output the estimated Click-through rate.The PDU model was also implemented under the Pytoch deep learning framework and tested on the Movielens-1M dataset.The GAUC value of the PDU model increased by 0.3%compared to the DIAR model,while the Logloss value decreased by 0.3%.