Research On User Portrait Recommendation Algorithm Based On Massive Retail Data

In the era of big data “Internet+”,the amount of related retail data has accumulated to more than 30 T.During the whole retail marketing process,it plays an important role to explore the potential value of data.Therefore,this paper takes sampling method to collect retail sales data and other related information targeted at more than 8 million retailers,and then store?process and mine these data.On this basis,a multilevel and multidimensional user portrait model is constructed.The improved FCM algorithm is used to cluster the user portrait,and the improved recommendation algorithm is used to design and implement personalized retail information recommendation system.The theory and technology mentioned in this paper are applied in the system to achieve personalized recommendations for retail customers.The specific research contents are as follows:Firstly,using the terminal information collection system gathers the data.Aiming at the problems of widespread distribution and clutter of more than 8 million retailers across the country,this paper takes on-site visit to collect retail sales data and improve the patrol plan.Considering repeated collections occur in the data collection area,the API positioning of Baidu is introduced in the system to improve the collection efficiency and optimize the collecting data quality.Secondly,via multi ways to perceive market information and analyze the data mining process.This paper integrated the terminal information collection system,the retail order system and other business data system.The integrated large-scale data is uniformly cleaned,integrated and converted,and the data dimensionality reduction model is established according to the data demand.It introduces the Spark distributed in Hadoop to deal with massive data sets and realizes data architecture for enterprise,besides,it shares business data among multiple systems etc.Thirdly,studying multilevel and multidimensional user portrait models.A user image index system is established for the on-demand classification of retail information,and a four-dimensional array is proposed to construct a multi-level and multi-dimensional user portrait model.The master model is divided into sub model of basic retail dimensions,sub model of retail domain dimension,sub model of retail management dimension and sub model of social business dimension.The improved FCM algorithm is used to cluster the multilevel user portraits.Comparing the group user portraits obtained by the improved FCM algorithm with that by the traditional FCM algorithm,the traditional K-Means algorithm and the improved K-Means algorithm,it shows that the algorithm of this paper reduces the number of iterations and the average time consumed,and the MAE has declined dramatically,besides,the accuracy and recall rate have promoted.Fourthly,this paper is base on the recommendation technology research of multi level user portrait model.On the basis of the group user's portrait,to solve the problem of the data with sparseness on collaboratively filter algorithm and the accuracy of similarity is not high,the fusion discrete quantity and user preference are proposed to correct the similarity,which brings more accurate similarity calculation matrix.Finally,a collaborative filtering algorithm based on the basic discrete quantity and user preference similarity is proposed to carry out personalized recommendation.Experiments results show that the algorithm mentioned above can achieve better prediction accuracy with extremely sparsity of data compared with other algorithms.Finally,analyze the effects of the personalized retail information recommendation system.In view of the poor quality services of retail information,the theory and research methods proposed in this paper are applied to the system,and the best results are actively pushed to users.The application results show that the multilevel user portraits and improved algorithm proposed in this paper have achieved remarkable results and can improve the quality of terminal data collection,besides when put it in the recommendation information of retail products and service across the country,the overall sales trend has been improved.The result shows that the growth rate has increased by 4.46%,and the customers have increased by 18.21%,of which second and third level customers have increased by around 50,000.