Research On The Recommendation Algorithm Under The Constrained Gale-shapley Mechanism

The matching mechanisms in many real two-sided matching problems do not always lead to the stable matching,and it usually can be manipulated.In the manipulable two-sided matching problem,it is of great practical significance to aid the agent making choice.Based on the classical Gale-Shapley mechanism,the constrained Gale-Shapley mechanism is proposed.There are two kinds of recommendation problems in different conditions.The first recommendation problem is to recommend to the agent which does not choose any other opposite anget.The second recommendation problem is to recommend to the agent which has chosen some opposite angets.In order to solve the problems of the two kinds of cases,the recommendation algorithm based on F-score and the recommendation algorithm based on the nearest neighbor association rules are proposed.Finally,the two algorithms are analyzed in the case of the parallel enrolment mechanism.1)The research on the constrained Gale-Shapley mechanism,and the two recommendation problems.Based on the classical Gale-Shapley mechanism,the constrained Gale-Shapley mechanism is proposed.There are two kinds of recommendation problems in different conditions.The first recommendation problem is to recommend to the agent which does not choose any other opposite anget.The second recommendation problem is to recommend to the agent which has chosen some opposite angets.2)The recommendation algorithm based on F-scoreIn order to solve the first kind of recommendation problem,the recommendation algorithm based on F-score is proposed.According to the historical matched data and the agent’s risk preference,the proposed algorithm calculates the agent’s standardized F-Score to each agent on the opposite side.The agent on the opposite side which has the biggest standardized F-Score is recommended to the original agent.3)The recommendation algorithm based on the nearest neighbor association rulesIn order to solve the first kind of recommendation problem,the recommendation algorithm based on the nearest neighbor association rules is proposed.Each selection list of the agent is considered as a transaction.All selection lists of the agent’ nearest neighbor consist the transaction data set.So we can carry out the association rules by mining on the nearest neighbor transaction data set.And the algorithm can recommend to the opposite agents according to the association rules and the selected opposite agents.4)the experiments to analyze the proposed recommendation algorithmsAs for the first kind of recommendation condition,whether the successful matching is the main consideration.so the paper use the success recommendation rate and the total utility as the two indicators to evaluate the recommendation algorithm based on F-score.With contrast experiments between the proposed algorithm and other two recommendation algorithms using the real students and college admission data,it is found that the proposed algorithm has a high success recommendation rate and can respond to the agent’s risk preference.For the second kind of recommendation condition,the selected opposite agents are the result of the balance of the agent’s conditions.so the paper use the success recommendation rate and the coverage as the two indicators to evaluate the recommendation algorithm based on the nearest neighbor association rules.With contrast experiments between the proposed algorithm and the recommendation algorithm based on association rules using the real students and college admission data,it is found that the proposed algorithm has a higher coverage than the recommendation algorithm based on association rules.