Research On Combining Multi-source And Heterogeneous Data In Product Sales Forecast And Delivery Strategy

In the context of "Internet+" the amount of data generated by large manufacturing companies expanding their business through PC,mobile and offline channels has reached the PB level from the TB level.While business decision-making brings new opportunities,it also poses new challenges to traditional data mining methods.Faced with the problem of data space redundancy and complex information from multiple sources and heterogeneity caused by the order data and sales data of more than 8 million customers in the global market,we need to study how to effectively integrate a large number of heterogeneous heterogeneous data sources.It also solves the problems of large backlog of social inventory and unbalanced development of products in the process of product launch,and provides decision-making basis for accurate product launch in the new form.This article aims at the problem of inaccurate business decision-making caused by the unclear relationship between multi-source heterogeneous subscription data and business process mapping in the process of product launch and operation.Research on key issues in data mining such as sales forecast and pilot customer recommendation,product launch strategy model during the launch process.The main work and contributions are as follows:(1)This paper constructs a method of customer information collection and processing based on user preferences in a multi-source heterogeneous environment.The data collection work is completed through the offline inspection,collection,and integration reporting methods of the regional commissioners and the way of Internet requests,and is imported into the distributed multi-source heterogeneous database.Aiming at the problems of inconsistent representation of multi-source heterogeneous data,multiple fields and redundant values,and multiple indicators,this paper proposes the use of XML technology and the establishment of a standard database of data attributes to eliminate the inconsistency of data dimensions,and uses the sorting method to detect and remove similar redundant data.At the same time,a data fusion model based on OWA operator and user preference is proposed to weaken the structural ambiguity and semantic difference of data information and improve the reliability of user decision-making.(2)Aiming at the problem of overstocking during the product launch process leading to the backlog of social inventory,this paper proposes to use regional sales forecasting algorithms that integrate multi-source heterogeneous data to forecast regional launches.Based on the existing market segmentation research,combined with the business district auxiliary factors,project customer order data to each business district,and increase the accuracy of forecasting by adding customer sales capabilities and regional consumption characteristics of the business district to model the factors.Using the customer order data projected to each business district,a product transfer matrix is proposed to describe the degree of influence of each business district on sales.On this basis,after reducing the dimensionality of the characteristic data that affects sales through the gray correlation analysis method,the XGBoost algorithm with a regularization term added to the objective function is used to reduce the complexity of the prediction model and sell the business circle market in the future time period.prediction.Experimental results show that compared with other algorithms,the improved algorithm can effectively solve the problems of poor prediction stability and low accuracy.(3)Aiming at the problem that the selected area is too large during the product launch process,which leads to unbalanced product development,this paper proposes a pilot recommendation for multiple areas using a customer recommendation algorithm that fuses heterogeneous data from multiple sources.Carry out a comprehensive value evaluation of customers across the country,and then use the subspace decomposition method to analyze the purchase situation of products in each region.Combine the customer value results and product purchase situation results to build a global user project score matrix,and calculate the coupling object Similarity recommends the best target customers as a product launch pilot.Experimental results show that compared with other algorithms,this algorithm can effectively alleviate the cold start problem of the recommendation system,and improve the robustness and accuracy of the recommendation algorithm to varying degrees.(4)Aiming at the problem of inaccurate delivery caused by too many deliveries and too large selection areas,this paper proposes a precise delivery strategy model that combines regional sales forecasts and pilot customer recommendations.A multi-factor weighted evaluation method is proposed based on individual customer characteristics to screen initial launch pilot customers,and Logitic regression models are used to determine the final launch pilot from multiple dimensions and indicators based on the sales trends of products in different business district areas.Established a product launch strategy model.Finally,four pilot cities in Zhejiang Province were selected to conduct product launch tests.The results show that the product ordering rate and sales trend after using the launch strategy proposed in this article are better than the original launch strategy of the manufacturing enterprise.In the launch practice,the market share and launch revenue Increased by 13.5%and 26.3%,respectively.