Research On High-speed Convergent Latent Factor Analysis Molde Based On Particle Swarm Optimization

With the explosive growth of the Internet,the information people need is contained in a large amount of data.The recommendation system can help people filter out useful information from these data,which can be represented by High-Dimensional and Sparse(HiDS)matrices.Because the recommendation technology based on latent factor analysis(LFA)has the advantages of high computational efficiency,low space complexity,and strong scalability,it has been widely studied and applied,and is an effective method for processing HiDS data.In the LFA model,the stochastic gradient descent(SGD)algorithm,as the most popular solution,is often used to extract latent factor(LF)in HiDS matrices.However,the performance of LFA model based on SGD is affected by the learning rate.In order to obtain accurate prediction results,a suitable learning rate needs to be selected,which requires a lot of time and energy.Therefore,in order to improve the performance of existing LFA models,this paper proposes a high-speed convergent LFA model based on particle swarm optimization(PSO)algorithm,and conducts research work around the optimized model.The main work completed is as follows:(1)A PSO-based LFA model is proposed.This model applies the rules of PSO to the LFA model,each particle is regarded as a learning step that needs to be optimized,and the learning step is automatically selected by updating the experience of particle itself and the experience of swarm,so that the learning step can be self-adaptive,thereby obtaining high computational efficiency and competitive prediction accuracy.This model is significantly superior to the traditional LFA model in terms of time cost.(2)A novel position-transitional PSO algorithm is proposed.For the problem that particles are easy to fall into the local optimal solution,the particles in the swarm consider the previous flight distance during each flight,that is,the velocity of particles during the previous iteration,so as to expand the scope of particle exploration,avoid the situation that particles fall into the local optimum and unable to jump out,and balance the global search ability and local search ability of particles.(3)An LFA model based on improved PSO is designed.This model applies the rules of position-transitional PSO to the LFA model.Through experimental comparative analysis on data sets generated by large-scale industrial applications,the results show that the model proposed in this paper is superior to other LFA models in terms of prediction accuracy and computational efficiency,which shows special efficiency in computational efficiency and greatly improves the performance of original model.