Research On Sparse Reconstruction And Personalized Recommendation System Based On Evolutionary Algorithm

In practical engineering and scientific fields,there are complex optimization problems.Due to their complexity,dynamic change,binding force,nonlinearity and difficulty of modeling,which put higher requirement for the technique of optimization and computing,traditional mathematical methods can't well solve this category of problems.A new heuristic algorithm has become a new research hotspot for solving these kind of problems and attracts more and more researcher's attention.Evolutionary algorithm is a kind of nature-inspired random search algorithms by simulating natural evolution.It shows the intelligence,versatility and global optimization ability,which enables it to solve various complex optimization problems.This thesis gives a thorough research on the application of differential evolution(DE)algorithm for sparse recovery and personalized recommendation based on multi-objective evolutionary algorithm.Firstly,this paper introduces the research background of the sparse recovery problem and the personalized recommendation.Secondly,this thesis analyzes the current research status of DE algorithm,multi-objective optimization algorithm and the application of evolutionary algorithm in optimization problems.Thirdly,this essay makes the application research of adaptively differential evolution algorithm on signal sparse recovery.Finally,this paper describes research work of personalized recommendation based on the multiobjective evolutionary.The main work of this paper is described as follows:1)When solving the sparse recovery problem,this paper proposes a local search enhanced differential evolutionary algorithm(LSE-ADE)for this problem.The main innovation of LSE-ADE is to apply the DE algorithm for solving the signal sparse recovery problem,as DE shows the strong global search ability.In addiction,an adaptive control strategy is further extended to solve sparse recovery in this paper.Multiple DE mutation strategies are adopted in our scheme and the preferred DE mutation strategies will be selected adaptively,which can suit the different phases of evolutionary process.Therefore,the natural limitation of single DE strategy is avoided,which can tackle the sparse recovery problems with various complicated features.Finally,a local search method(i.e.,STM)is embedded into the evolutionary process of DE,which combines the advantages of fast suboptimal convergence speed by STM with DE,in order to give a good comprehensive performance in solving the sparse recovery problems.2)When solving personalized recommendation problem,this paper proposes an enhanced personalized recommendation based on multi-objective evolutionary aglorithm(MOEAERS).The main improvements are described as follow.Firstly,MOEA-ERS proposes a new estimation method to measure the recommmendation accuracy,which can correctly evaluate the recommender accuracy and thus get a higher recommendation accuracy;Secondly,MOEA-ERS introduces an extreme-value-guided operator into three objective functions(i.e.,accuracy,coverage,novelty),which can effectively guide the direction of evolution to the positive or the better direction.Therefore,the proposed algorithm improves the recommendation accuracy,coverage and novelty simultaneously...