Personalized Logistics Service Recommendation Based On Non-Negative Matrix Factorization

With the continuous development of "Internet plus logistics",the third-party logistics distribution service has become the best choice for most small and medium-sized enterprises to reduce the logistics distribution cost and improve the quality of logistics distribution service.In order to break the information barriers between the provider and demander of logistics service resources,the relevant government departments and enterprises in China have invested in the construction of a number of logistics information service platforms.As one of the ways for small and medium-sized enterprises to choose third-party logistics distribution services,platform can promote the integration of logistics service resources and information sharing to a certain extent,and improve the efficiency of logistics distribution.However,on the logistics information service platform where the provider and demander of service resources gather,the number of logistics distribution services is large and the actual quality is unknown,different customers also have different preferences on the service's timeliness,safety and other attributes.Therefore,it is difficult for the small and medium-sized enterprises with personalized demand for logistics distribution services to efficiently choose services that meet their own needs,resulting in low user satisfaction.For the above purpose,this paper studies the logistics distribution service recommendation method that can fully tap the personalized needs of users.It mainly builds the personalized logistics distribution service recommendation model and gives the hybrid recommendation algorithm(MLEW)to solve the model.Firstly,the scoring data of each attribute of the service are preprocessed to obtain the corresponding scoring matrix.Secondly,a feature extraction method based on the projection non-negative matrix decomposition was proposed to solve the problem of highly sparse scoring matrix,and the iterative algorithm was given to obtain the coefficient matrix containing user preference information,and then the user similarity matrix was obtained,based on the user similarity matrix,similar user sets of target users were found.Then,based on the similar user set obtained,the weighted Slope One algorithm is used to predict the initial rating values of each attribute of the service by target users,and then the information entropy is used to correct the initial values.Finally,the weighted sum of the predicted score values of each attribute by the modified users was carried out to get the final predicted comprehensive score value,and according to the predicted comprehensive score results,the results were recommended by TOP-N recommendation method.In the end,this paper conducts numerical experiments based on the user rating data set generated by the real logistics distribution service information of Alibaba logistics service platform.The experiment is mainly divided into MLEW algorithm performance analysis and model prediction results analysis.Experimental results show that the precision rate and recall rate of MLEW are 4.31%and 3.1%higher than WSO algorithm,and 1.89%and 2.85%higher than LEW algorithm,respectively,in the data set of this paper.In addition,user satisfaction was tested according to the score prediction results,and the results showed that user satisfaction could be 9%higher than that before the recommendation using MLEW algorithm.Therefore,the personalized recommendation method proposed in this paper can effectively improve user satisfaction,thus promoting the virtuous cycle development of platform,service resource provider and demander.