Research On Fairness Recommendation Algorithm Based On Weighted Meta-path

In recent years,the development of the Internet has greatly increased the scale of data and information.In this case,the recommendation algorithm is particularly important.Recommendation algorithm can help people quickly find the information they are interested in and save the time of users.Now recommendation algorithm has been very mature,and many of them have been successfully applied to the industry and achieved good results.However,most of the current research algorithms only focus on improving the accuracy of recommendation,thus ignoring the interpretability of recommendation,and there is little research on the fairness of recommendation.Therefore,this paper makes an in-depth study on the recommendation system from these two starting points.First of all,aiming at the problem of poor interpretability in recommendation algorithm and the problem of sparse data in current recommendation system,this paper proposes a recommendation method(Reinforcement Learning-Heterogeneous Information Network,RL-HIN)which combines reinforcement learning and heterogeneous information network.Firstly,the method analyzes and selects the reasonable meta path,and then uses reinforcement learning network to train the weight of meta path,Finally,the best weight of meta path is obtained for recommendation.This method uses heterogeneous information network,effectively solves the problem of data sparsity,and uses reinforcement learning to select the optimal weight of the meta path,which not only greatly improves the accuracy of the recommendation,but also increases the interpretability of the recommendation,so as to facilitate the follow-up research of data mining.Secondly,in view of the fairness problem in the recommendation system,this paper abandons the traditional means of recommendation,and proposes a fair personalized recommendation method.This method separates the hot item recommendation from the non hot item recommendation,and carries out the recommendation according to the pre agreed recommendation length and the proportion of the hot item recommendation and the non hot item recommendation,while ensuring the accuracy and fairness of the recommendation.Thirdly,in view of the problem that the non popular items cannot be recommended due to the lack of data,this paper designs the fair and accurate function of non popular items,transforms the recommendation problem existing in non popular items into the problem of finding the best combination of items,and then completes the recommendation of non popular items,finally solves the problem that non popular items are ignored in the recommendation,and achieves the goal of fair promotion Recommendation effect.Finally,the performance of the proposed method is verified and compared in the real Movie Lens 1M data set,and the accuracy and fairness of the algorithm are proved by using RMSE,MAPE,Diversity,Fairness,Accuracy and Recall.