Marketing Analysis In Multi-Dimensional Space Based On POS Data

Marketing analysis plays an important role in retailing, and data warehouse, used to store historical sales data, has been widely applied at present. With the rapid growth in the amount of retailing data, the level of data management and marketing analysis methods are different and unbalanced development among corporations.The foundation of this paper is MVS Retailing Business Intelligence Solution (MRBIS). The data format and content of cash tape (also called POS data) are substantial unified in each corporation, making effective use of it can provide marketing decision support. MRBIS builds the unified business intelligence (BI) architecture in retailing based on POS data. The design process of MRBIS and marketing analysis in multi-dimensional space using POS data are discussed in this paper. The main work is as follows.(1) Study and establish Unified BI architecture based on POS data in retailing to provide data management and support for marketing analysis. Analyse subject areas and logical model and then design multi-dimensional DW, study the implementation of ETL and handle the procedure of loading data from several heterogeneous databases into the DW, discuss the design methods of dimensions and measures according to the character of POS data, and build multi-dimensional datasets.(2) Carry out marketing analysis using multi-dimensional datasets of the unified BI architecture. To have a thorough understanding of sales situation, adop OLAP query to obtain statistical information of commercial equation, tickets breakdown and category structure of baskets. To get the character of customer preference, adopt cluster analysis in behavior segmentation of customers. To provide suggestion for binding promotion and to optimize shelves management, adopt association rules mining in association analysis of products. Take advantage of real POS data in retailing from MRBIS. this paper provides horizontal and vertical comparison between k-means algorithm and EM algorithm, the result shows that EM algorithm has a higher stability and classification effect. Discuss the parameters of association rules based on the character of POS data. and then apply association rules analysis in different category levels of products. (3) Conduct detailed discussion about the problem of object promotion to promote a specified product using multi-dimensional datasets. One product has no vantage in global space, after the decomposition of the global space, this product maybe stand out in several subspaces. Object promotion is used to provide evidence for promoting a specified product by finding dominant subspaces. On the basis of proposed models, one is a model called multi-dimensional space and the other is called multi-dimensional region, this paper puts forward multi-dimensional and multi-level space model, whose solution space has a higher complexity. The solution of this model adopts materialization strategy based on fitting method to solve the problem. A cube called LR-Cube and a query algorithm called LRAPQuery are proposed, this strategy guarantees completeness, uniqueness and high efficiency in subspace traversal, the result of experiments shows that LRAPQuery based on LR-Cube ensures high efficiency and accuracy.(4) Implement MVS Retailing Business Intelligence Solution. Implement the BI architecture of MRBIS using SQLServer database, utilize Data Engine Service to implement the design of data warehouse, Integration Service to implement the ETL process, Analysis Service to implement the design of multi-dimensional datasets and data mining models, Reporting Service to create reports.