Research On Recommendation Algorithm Based On Multi-objective Optimization And Preference Strategy

The continuous development of the Internet has brought great convenience to people,but at the same time,it has also brought the problem of information explosion,how users quickly find the information they need from the mass of information,how businesses recommend their products to users have become an urgent problem to be solved.Therefore,recommendation system as a tool to solve the information explosion has been applied in various fields.With the continuous development of recommendation system,the traditional recommendation system which takes accuracy as the only evaluation index can not meet the needs of people,and people put forward new requirements for recommendation system,which are represented by diversity and novelty.This paper studies the existing recommendation system based on multi-objective optimization,proposes two new objective functions to solve the problem of unreasonable objective function setting of the original algorithm,and introduces a preference strategy to solve the problem of unreasonable objective function weight setting.In addition,aiming at the problems of poor distribution and convergence of the original algorithm in the three-objective space,a new multi-objective recommendation algorithm is proposed to improve the performance of the algorithm.The specific research work includes the following three aspects:(1)Based on the multi-objective optimization algorithm based on clustering,this paper proposes two objective functions for user clustering to balance the diversity,novelty and accuracy of the recommendation algorithm.A new method is proposed to dynamically determine the number of nearest neighbors,which ensures that the algorithm adaptively selects the nearest neighbor users according to the number of users in the cluster.In addition,the initialization method based on the proportion of similarity and diversity of the initial cluster was applied to ensure the similarity and diversity balance of the population during initialization.Finally,the diversity and novelty of the recommendation system are improved without sacrificing more accuracy.(2)This paper proposes a recommendation algorithm based on improved resource allocation and preference user guidance.By introducing user similarity information and preference information to modify individual fitness function,the weight of accuracy objective function is increased.At the same time,in the crossover stage of multi-objective optimization,a preference user-guided multi-parent crossover method is adopted,which makes the algorithm generate children based on accuracy preference during crossover and guides the evolution direction of the algorithm.The results show that the accuracy and diversity of the algorithm are improved.(3)This paper proposes a three-objective recommendation algorithm based on mixed preference strategy.In the multi-objective optimization stage,the algorithm is optimized by NSGA-?,and the distribution of the algorithm is improved by the individual retention mechanism of NSGA-?.In the initial stage,a hierarchical initialization method based on item degree was adopted to ensure the distribution of population on the objective function of novelty.In addition,a hybrid preference strategy based on NSGA-? is used to design the crossover operator of the algorithm,and the crossover mode is dynamically adjusted by the crowding condition of the population distribution.The results show that the distribution of the algorithm and its performance in the accuracy objective function and novelty objective function have been greatly improved.