Research On Personalized Recommendation Model Of Multi-service System Integrating LightGBM And DeepFM

With the construction and development of China’s government Big data platform and digital government,the problem of information overload of the government Big data platform multi business system has become increasingly prominent.The government Big data platform needs to make more effective use of the accumulated user historical behavior data to achieve accurate estimation of the political services that users may need,thus improving the practicability and convenience of the government multi business system.Therefore,it is of great significance to attach importance to and solve the problem of information overload for the development of China’s government Big data.However,when the government Big data platform uses the Click-through rate prediction model to estimate the government services that users may need,there are many problems,which are mainly manifested in the following three points: first,the Click-through rate prediction model of the model has poor robustness and needs to be improved in accuracy;Second,model space consumption is large,and data dimension explosion is easy to occur when processing data;Third,the model is mainly used to estimate the Click-through rate of online advertising,which is not suitable for the needs of the multi business system scenario of the government Big data platform.This paper takes the personalized recommendation needs of the multi service system of the government Big data platform as the background,analyzes the actual characteristics of the government multi service system,and proposes a highly accurate and robust integration LightGBM and DeepFM Clickthrough rate prediction model LGDF,which is used to estimate the probability of users clicking on a government service.This article mainly focuses on the following three aspects:1.In view of the data characteristics of the government Big data platform scenario,the LGDF proposed in this paper is improved on the basis of LightGBM and DeepFM algorithms.On the premise of not deleting the original data,it proposes the method of splicing the integer leaf value vector generated by LightGBM with the original data line,and innovates the data usage between models.Compared with traditional methods,this data usage method reduces data dimensions,saves memory space,and improves the estimation efficiency and accuracy of the model.2.In the model experiment part,use the public performance evaluation dataset Criteo commonly used in the Click-through rate prediction model to train and verify the accuracy of the model.After reasonable preprocessing of the dataset,different hyperparameters are set on the model to examine the impact of different parameter values on the model,and ultimately obtain the optimal hyperparameters and model performance.In order to confirm the accuracy of the model,the LGDF model proposed in this paper is compared with the traditional mainstream Click-through rate prediction model,and the results show that the AUC value and Log Loss value of the LGDF model proposed in this paper perform better.3.Design and implement a personalized business simulation recommendation prototype system based on the prediction algorithm proposed in this article.The system integrates functions such as model parameter setting,model training process visualization and effectiveness evaluation,and display of prediction results.Research and demonstrate the feasibility and effectiveness of the proposed model in business recommendation.