Research On Interpretable Recommendation Algorithm For Diversity Requirements

With the rapid growth of Internet-scale leads to information overload,which leads the user can not get to the product they are interested in.The emergence of the recommendation system has effectively solved this problem,but the current system is still facing other personalized needs of users,and some users are paying more and more attention to the detailed process and steps of the system recommendation.How to improve the accuracy of recommendation algorithms,the diversity of recommendation results,and enhance the interpretability of recommendation results have become the focus of this research.The diversity-oriented privacy protection recommendation method,and the other is the diversity-oriented GBDT interpretability recommendation method.These methods increase the diversity of methods while maintaining the accuracy of the system.The interpretability and readability of the method are enhanced by combining relevant interpretability techniques.This research proposes a diversity-oriented privacy protection recommendation algorithm.In order to increase diversity,the idea of this method is to enrich users' data information in other fields by adding multi-source data sets.Based on the differential privacy and implicit semantic model,an enhanced neighborhood privacy protection method is proposed,which can effectively protect user privacy data.This algorithm fully combines the characteristics of multiple data sets,effectively improving diversity and accuracy.The results indicate the effectiveness of the proposed method.An interpretable recommendation algorithm for diversity-oriented GBDT is proposed.This algorithm combines the logistic regression method and the SHAP method based on GBDT for advertising recommendation.This algorithm contains the following advantages.First,the features are automatically screened and combined by the GBDT model to generate new discrete features,which serve as the input of the LR model.At the same time,the model integrates a variety of different features such as users,items,and context,effectively improving the diversity and accuracy of recommendation algorithms.Not only that,but the effectiveness and interpretability of the algorithm are improved because of the visualization of models and features by using the feature important method of SHAP and GBDT.Compared with the original GBDT,the accuracy and diversity have been improved.