Information Core Optimization Based On Evolutionary Algorithm And Clustering In Recommender Systems

As an efficient method for alleviating information overload,the recommendation system has won praise from users.The collaborative filtering algorithm makes personalized recommendations to users based on users' historical information,which is applied widely because of its effectiveness and simplicity.However,noisy users and sparse data may limit the performance of the algorithm.In order to alleviate the above problems,the idea of information core users was proposed.The existing methods construct information core,however,just based on some pre-defined criteria.To address this issue,we model the problem of searching information core users as a combinatorial optimization problem and use evolutionary algorithms and their variants to search for information core users.Three algorithms of searching information core users are proposed as follows:(1)An evolutionary algorithm with elite population is presented to search for the information core users,where an elite population with ordered crossover is used to accelerate the evolution.Experiments validate the effectiveness of our proposed algorithm.Results show proposed algorithm is able to effectively identify information core users,leading to better recommendation precision compared to several existing heuristic strategy methods and the conventional collaborative filter.In addition,algorithm is shown to significantly reduce the time of online recommendation.(2)An evolutionary algorithm based on variable-length encoding is proposed to alleviate the disadvantages of the slow convergence rate of traditional evolutionary algorithms when the system contains large-scale users.First,the population is initialized where the population's length is set beforehand.Then excellent users are combined through the crossover operator that combines the gene in individuals.The length of the individual is increased or decreased through a specific mutation operator and some of the genes in the individual are changed by the mutation operator.Furthermore,through the strategy of resetting the population,the solution space is extensively searched.The experimental results show that this method is superior to the evolutionary algorithm with elite population,and it can find fewer information core users which are more excellent much faster,which reduces the time of online recommendation.(3)The concept of virtual information core user is proposed.The virtual information core is dedicated to solving the problems caused by sparse data and excessive number of users.Firstly,it clusters users according to the users' historical data,and then virtualizes users belonging to the same class into one virtual user to reduce the sparseness of the data.Finally,an evolutionary algorithm with elite population is used to search for virtual information core users among virtual users.The experimental results show that the proposed algorithm is greatly improves the two indicators called precision and recall.And it also alleviates the problem of scalability in some extent.