Research On Many-objective Recommendation Algorithms Considering Multistakeholders For E-commerce Platform

Traditional recommender systems recommend items according to users’ needs and preferences,but users are not the only participants in the recommender systems.In the multisided marketplace of e-commerce platforms,the needs of other stakeholders are equally important,i.e.,providers and the platform.Suppose a recommender system only pays attention to the preferences of users and ignores the needs of providers.In that case,providers will give up using the recommender system or turn to other platforms.The loss of providers will reduce the type and quantity of items and affect the platform’s profitability.Therefore,a comprehensive recommender system should incorporate the objectives of multiple stakeholders into the recommendation process and provide strategies that meet the preferences of multiple participants.For providers,provider coverage is currently used to measure the quality of recommendations.However,only considering provider coverage may lead to the following problems: on the one hand,the items in a user list may come from a limited number of providers,i.e.,the distribution of providers is lack of diversity;On the other hand,a provider may only be recommended to a limited number of users,resulting in a low frequency of recommendation by the provider.Both of the above situations will lead to the imbalance of provider recommendations.Additionally,there are many problems in the process of solving multistakeholder many-objective recommendations: for one thing,the commonly used linear weighting method cannot get the optimal solution because there are often different degrees of conflicts between multiple objectives;for another,when the traditional crossover operator is used to optimize many-objective models,the crossed chromosomes tend to generate infeasible solutions,which requires very complex repair operations.In addition,the many-objective recommendation model generates a fixed number of recommendation lists for all users of the e-commerce platform at one time,therefore the dimension of the solution space is quite large,resulting in low convergence of the generated solution.Therefore,the research on many-objective recommendation algorithms considering multistakeholders for e-commerce platforms has important theoretical and practical significance.The main research work and innovations of this thesis are as follows:(1)A similarity model based on customer lifetime value and user preference is constructed.The model considers the user ratings of items and the users’ long-term transaction characteristics.The user-based collaborative filtering algorithm uses this model to calculate user similarity,which significantly improves the accuracy of item recommendations in user recommendation lists.Based on the initial recommendation lists of users,a multistakeholder recommendation algorithm considering customer lifetime value is designed.In the process of solving the algorithm,the N-block heuristic crossover operator is constructed,which solves the limitation that the traditional crossover operator tends to generate infeasible solutions which do not meet the chromosome coding constraints.The optimized solutions not only ensure high recommendation accuracy but also improve diversity,novelty,provider coverage,and platform profitability.(2)A provider visibility utility model is proposed.The model is composed of provider coverage,user reach coverage,and provider entropy.It solves the problem that the provider coverage does not consider provider distribution balance,and lays a foundation for the establishment of the many-objective recommendation model.In addition,in the process of many-objective optimization,only considering the provider visibility utility model can simultaneously optimize the provider coverage,the frequency and diversity of provider recommendations,which can significantly reduce the complexity of many-objective optimization.(3)Considering the four objectives of users and providers(recommendation accuracy maximization,recommendation diversity maximization,recommendation novelty maximization,and provider visibility maximization),a cascaded hybrid recommendation algorithm considering multiple stakeholders is designed.The algorithm constructs a multiparent probabilistic heuristic crossover operator,which not only considers the frequency of genes in multiple parent chromosomes,but also the offspring chromosomes generated after chromosome crossover meet the chromosome coding constraints and do not need repair operation.The solutions optimized by multiparent probabilistic heuristic crossover operator have better diversity than the traditional crossover operator and improve the convergence of the generated solutions.