Research On Hybrid Recommendation Model Based On LFM

Purpose—In the research of recommendation system,the problem of improving recommendation accuracy and reducing data sparsity has always been a research hotspot in this field.Iterative optimization and matrix decomposition in latent factor model reduce data sparsity.Compared with the batch gradient descent method,the stochastic gradient descent method in the process of iterative optimization has a significant improvement in efficiency.However,the noise analysis and convergence of the stochastic gradient descent method are worse than the mini-batch gradient descent method in the case of large data.After comparing the stochastic gradient descent method with the mini-batch gradient descent method,the mini-batch gradient descent method is proposed to replace the stochastic gradient descent method in latent factor model.In terms of improving recommendation accuracy,the recommendation accuracy of hybrid recommendation model is generally better than single recommendation model.A hybrid recommendation model based on neighborhood and latent factor model using mini-batch gradient descent method is proposed.The hybrid model is KLFM hybrid recommendation model.Mixed methods are three fusion strategies based on linear weight allocation: equal weight fusion method,effective weight fusion method,particle swarm optimization and weight optimization combined PSO optimization weight fusion method.Design/methodology/approach—In terms of noise analysis and convergence,the stochastic gradient descent method in traditional latent factor model is compared with the mini-batch gradient descent method.This paper proposes equal weight fusion method,effective weight fusion method and PSO optimization weight fusion method.Using the above three fusion strategies,the latent factor model of the mini-batch gradient descent method and the neighborhood-based recommendation algorithm will be mixed,and the recommendation accuracy of the mixed recommendation model will be measured by comparing it with the single recommendation model in MAE,MAPE and RSME.Findings—Compared with the stochastic gradient descent method,the mini-batch gradient descent method has lower oscillation in the noise analysis figure,and the iteration process is relatively stable,and it is easy to determine whether it converges or not.According to the experimental data,it can be concluded that the mixed recommendation model is superior to the single recommendation model in the accuracy evaluation indexes of MAE,MAPE and RSME.Among the three fusion strategies based on linear weight combination,the PSO optimization weight fusion method is superior to the other two fusion strategies in the three indexes of MAE,MAPE and RSME.Limitations of research/Implications—In the process of model mixing,less recommendation models are selected;more choices can be made in the fusion strategy;more evaluation indexes can be selected to evaluate the recommendation model from different angles and in all directions.Actual impact—In large data volume,using mini-batch gradient descent method is better than stochastic gradient descent method in iteration accuracy and convergence;in the study of KLFM hybrid recommendation model,the effect of KLFM hybrid recommendation model is better than single recommendation model;in the comparison experiment,PSO optimization weight fusion method is used to evaluate the accuracy of mixed KLFM hybrid recommendation model in MAE,MAPE and RSME.The performance of the index is the best,and the mixed effect of the mixed model is the best.Practical implications—An optimization method based on latent factor model for iterative optimization in large data environment is proposed.In latent factor model,mini-batch gradient descent method is used instead of stochastic gradient descent method.Based on the latent factor model of mini-batch gradient descent method,three linear weight fusion strategies are used: equal weight fusion method,effective weight fusion method,PSO optimization weight fusion method combining particle swarm optimization and weight optimization.The neighborhood-based recommendation algorithm is mixed with KLFM hybrid model.