Research On Recommendation Algorithms With Constraints On Item Capacity

As one of the most effective information filtering methods in the era of big data,the recommendation system has been deeply studied and widely used in academia and industry.A complete recommendation system generally has three main participants:users,item suppliers,and recommendation system operators.Traditional user-oriented recommendation algorithms are usually difficult to avoid the problem of item popularity bias,which greatly harms the fairness of item recommendation requirements of item suppliers,the other main participant in the recommendation system.This article considers the different needs of item suppliers and users at the same time,and carries out the recommendation algorithm research for the fairness and accuracy of recommendation.The main research content includes the following three parts:(1)Propose the item capacity restriction plan.In the recommendation results generated by common recommendation algorithms,the proportion of popular items recommended is much greater than the proportion of their interactions in historical records.This popularity bias will produce the Matthew effect of "the rich get richer" after multiple rounds of recommendations.Therefore,an item capacity limitation scheme is proposed,so that the number of times that each item is allowed to be recommended is proportional to the popularity of the item,thereby transferring part of the exposure opportunity from popular items to niche items,and improving the fairness of item recommendation.(2)Propose a user-item matching algorithm based on a greedy strategy in the context of item capacity restriction.Due to the limited item capacity,traditional Top N recommendations are no longer applicable.In this paper,a heuristic algorithm based on RSP greedy strategy is proposed to solve this constraint(item capacity limitation).In a single RSP algorithm,the recommendation accuracy has a large loss,so two post-processing methods are proposed,Post-Normalized and Post-Degree.Through analysis and experiments,the combination of Post-Degree RSP algorithm makes it have better performance in solving small user recommendation accuracy,the overall recommendation accuracy is guaranteed,and the fairness of item supplier recommendation is to cover the fairness index It also got a very good performance.(3)Propose a recommendation model based on the minimum cost and maximum flow(MCMF)for solving the item capacity restriction scenario.Because the recommendation result obtained by a single RSP algorithm itself is often not the optimal combination recommendation,the MCMF model is introduced and used for recommendation under the limitation of item capacity to solve the optimal combination recommendation result.Experiments show that the recommendation results solved by the MCMF model are also guaranteed in recommendation accuracy,and very good results are also obtained in the fairness of item supplier recommendation,that is,the coverage fairness index.In addition,compared with the P3 type recommendation algorithm,the MCMF method does not introduce any parameters,and has a comparable effect with the P3 type algorithm in the recommendation accuracy,and has an absolute advantage in coverage fairness.