Statistical Analysis Of Post Click Through Rate Based On Feature Optimization

Now more and more people rely on Internet recruitment platforms to find jobs.The Internet recruitment platform recommends suitable positions for job seekers through the position recommendation system.The position recommendation system sorts the positions through several important indicators,and the position click through rate is one of the important indicators.Therefore,the prediction of job click through rate is the key problem for the normal operation of job recommendation system.The position click through rate dataset contains user information,position information and interaction information between users and positions.The original data can further extract low-order interaction features,high-order interaction features and user behavior sequence features.With the emergence of user behavior sequence features,the commonly used feature extraction models are not comprehensive in feature extraction.Therefore,based on the characteristics of job click through rate dataset,this paper designs a job click through rate prediction model based on feature optimization.The model is divided into two parts: feature extraction model and prediction model.The feature extraction model incluedes: 1.use Light GBM model to extract low-order interactive features;2.use DNN model to extract high-order interactive features;3.use word2 vec model to extract the characteristics of user behavior sequence;These three features are combined into assembly features.Prediction model is a widely used logistic regression model in the industry.In terms of empirical analysis,this paper uses a recruitment platform’s job click through rate dataset,and the evaluation indicators of the model are AUC and logarithmic loss.Finally,this paper compares the model with the prediction model based on other feature extraction models.The experimental results show that the position click through rate prediction model based on feature optimization is better than other models;The prediction model with interactive feature extraction performs better than the prediction model without interactive feature extraction;The prediction model that extracts the characteristics of user behavior sequence performs better than the prediction model that does not extract the characteristics of user behavior sequence.