Research And Implementation Of Search Advertising Click Through Rate Prediction Algorithm

Internet advertising is mainly divided into search advertising,display advertising,etc.Among these,search advertising is the largest and fastest growing advertising delivery method.Search advertising is based on the query keywords provided by users,and it is the form of advertising that best understands users' search intention.The most critical technology for search advertising is the prediction of ad CTR.The prediction of advertisement click through rate is the prediction and analysis of the click rate of users' behaviors on the advertised advertisements.Advertising click rate is related to the order of advertising and advertising click fees and other factors.Therefore,the study of advertising click rate is of great significance to the improvement of search advertising revenue.The search advertisement utilizes the technology of the search engine,analyzes the query intent input by the user,extracts the corresponding keyword,performs semantic relevance calculation according to the extracted keyword,indexes from the advertisement database,and provides an advertisement search corresponding to the user's needs.result.When the search engine,such as the search engine and other platforms,has higher correlation with the key words input,the more interested the user is,the higher the possibility of clicking,the higher the click rate and the higher the advertising revenue.The main contents of this paper are as follows:(1)First,the data set is cleaned and pre-processed,statistical data information is extracted,shallow features are extracted,and common machine learning methods are used for feature engineering.The logistic regression model is used as a baseline method.As a baseline method,the logistic regression model takes the predicted click rate as a classification problem.In the traditional machine learning method,the combination of GBDT model and logistic regression model is used in the traditional machine learning method to combine features to further excavate the factors affecting the click rate and improve the nonlinear learning ability of the model.For data sparsity and lack of data,the factor decomposer FM algorithm,which has better results in various competitions in recent years,is used to optimize the results compared with the baseline method.(2)The similarity between the user query request and the advertisement title in the search advertisement plays a vital role in the click rate.This paper first uses the convolutional neural network model to extract deep-level similarity features,and to use it with artificial.The features extracted are combined and input into the prediction model of click-through rate.The experimental results are obtained and the effect is improved.(3)Recurrent neural network is more suitable for processing data,so this paper also studies the feature extraction based on the similarity of recurrent neural network.Experimental analysis and comparison of the experimental results of different types of recurrent neural network models,such as long and short term memory unit network,gate control loop network,and corresponding two-way network form,combine static attention and dynamic attention mechanism to compare the experimental results,extract the best phase similarity,so as to optimize the final result.