Analysis And Mining In E-commerce Site’s Sales Data

With the rapid development of e-commerce, the competition among enterprises be-come more and more fierce. The homogeneity of e-commerce websites’ goods is stron g-er, in addition to the factor of price, effective analysis and data mining, and putting forward the corresponding strategies and measures are important to the development of corporation. In this thesis, the bibliometric methods are applied to study the e-com merce, which has a certain reference value for researches of the e-commerce from the perspective of LIS field.Under the above background, based on the real e-commerce sales data, the thesis adopted quantitative method, comparative analysis approach, exploratory research meth od and empirical approach and so on to do data analysis and data mining to the sales data from the macroscopic and microscopic aspects. Relationship between the two par ts is progressive, as follows.The first part of thesis was to describe the overview of the sale of wireless term inal in different regions, and explore and verify the 80/20 Rule, long tail and power law distribution in the e-commerce from the macroscopic aspects. Wireless terminal sales area currently concentrated on the economically developed provinces. The 80/20 Rule meant the contribution of primary goods and general goods to the whole sales satisfied the 80/20 principle,20% of the category the main commodities contributin g nearly 80% of total sales, while the number of types about 80% of general goods contributing only about 20% of total sales. Long tail effect in e-commerce referred to the large-scale low level of Saturn Members and Copper Shield Members that occ upy over 50% of the total members accounted for about 50% of the total sales. In different regions between the Members’ contribution to the total order and the total a mount, we also found a similar situation that members located in the general sales t erritory contributed order quantity more than 48%, sales more than 46%. The long ta il effect above inspired decision-makers not only focused on loyal customers and maj or sales regions, but also concerned about the general customers and general sales ar eas located belonging to long tail. Providing that each customer is like a word in te xt, in analogy with Zipf’s law, the power-law distribution of e-commerce’s sales data was to explore whether there exist power law among customers’orders. Using Matl ab software, adopting the maximum likelihood estimation method and KS test to vali date users’orders, we found that customers’orders of all the goods and 3C represent ative product were consistent with the power-law distribution, but power-law index w as small. Through statistics on the top 20 customers and ranked last 10 customers’o rder, we found customers’ order network no special obvious scale-free network, but c onforming to the characteristics of long tail.The second part of thesis explored and validated the relation between the custome rs to purchase goods or goods category, and mined the association rules. By means of Weka software, the thesis adopted the FP-Growth algorithm to mine the association r ules to the book product,3C representative product and category, three sets of sample e-commerce sales data, according to the minimum support and confidence, leverage, c onviction, lift and other indicators, we analyzed and filtered the mined strong associati on rules from the theory, and changing the product number corresponding to the actua 1 goods, combined with the actual situation carried on the correlation analysis, some p ractical association rules were obtained. And due to the large number of customers an d the huge number of commodities, we did many experiments to get the desired assoc iation rules and shall set a lower minimum confidence.