M2M Recommendation Via Group Multi-Role Assignment

In order to solve the problem that rationally assigning the limited recommendation resources of the company,recommending the right items to the exact customers,and helping the company to get as much profit as possible,this paper is designed to concentrate on the study of the many-to-many recommendation problems under the limit of the recommendation resources.Recommendation is an important marketing approach of companies.According to the method of recommendation,the recommendation problems can be divided into two kind of recommendation problems,the recommendation problems with resource limit and the recommendation problems without resource limits.Actually,the recommendation problems with resource limits are usually the many to many recommendation problems.One customer can participate in many items,and one kind of items can be offered to many customers.Meanwhile,considering that different items have different profit margin and different customers have different purchasing power and matching degree to different items,how to rationally assign the limited recommendation resources and help the company to get as much profit as possible can be very complicated and important.In the real-world scenario and the problem analysis,this paper takes the real-work scenario that one financial company recommends its financial products to its VIP customers as a breakthrough point,and conducts the following studies and analyses: 1)Using the Role-Based Collaboration theories and the E-CARGO model framework to formalize the problem,abstracting the main element of the problem and establishing corresponding quantitative indicators.2)Using purchasing power,matching degree and profit margin to learn the optimal demarcation parameters for each product,which are used to divide the customers into three levels.Then appropriately choosing different method to fit different level customers' profit contribution index and matching degree index into the qualification index.3)Proposing two algorithms to solve the many-to-many recommendation problems under the resource limits.One is the ap-GMRA algorithm which combines the greedy strategy and the Group Multi-Role Assignment,the other is op-GMRA algorithm which combines the optimization strategy and the Group Multi-Role Assignment.The simulation experimental results show that,the proposed formal modeling and algorithms are both effective and practical.In terms of revenue optimization performance,the actual revenues caused by those two algorithms both can reach about 83%-84% of the prefect revenue,which is the revenue on the premise of knowing whether the customers will participate in the products or not.At the same time,the scale experimental results also show that the running time of the both algorithms are fast and efficient.By comparing the results,we can find that both of the algorithms have its advantage and disadvantage.In respect of the running time,the ap-GMRA algorithm is better than the op-GMRA algorithm.But in respect of the revenue optimization performance,the op-GMRA algorithm is better than the op-GMRA algorithm.Therefore,this paper offers some practice strategies and suggestions for different applications and scenarios.