Research On Segmentation Of Online Retail User Based On Improved ML-KNN Algorithm

With the development of online retail market,the market has turned from the seller-oriented turned to buyer-oriented.On one hand,online retail competition has become more and more hot and the cost for getting new customers is increasing quickly.On the other hand,consumers can get the price comparison information more and more easily.The online retailers were dependent on price and the sales growth and profit growth before.But it's harder for online retail sellers to keep the old customers now.Therefore,it is important for online retail sellers to carry out further user segmentation of customers.So this paper put forward a more accurate marketing strategy targeted identification of user characteristics,which suits the online retail sellers.The paper applied an 100,000 order data from the actual historical order data of online retail shop,which is a random sampling of over 10000,000 order data.First,through the consumer's purchase times,the paper evaluates whether the consumer is repeat-purchase consumer.Secondly,based on the discount amount,the number of the discount amount,the amount of payment and the number of consume,the paper applies the K-means data mining method to evaluate whether the consumer is price sensitive.In the end,based on the literature review,the paper adopts the RFM model which the time distance for the consumer so far as R,the total consumer number as F,the total consumer money as the M to evaluate whether the consumer is loyal.So the K-means method is applied too.There is correlation among the three labels.Loyal consumers tend to be more prone to repeat purchase,and price-sensitive consumers'loyalty will be relatively low,but price sensitive consumer are easier to do the repeat purchase.Therefore,in the perspective of data mining analysis,the paper put forward some suggestions for the retail sellers.And the paper reviews the multi-label learning algorithms and the RFM models.As the ML-KNN algorithm is first-order multi-label learning algorithm and lacks the considering of class imbalance.Two point improve suggestions are put forward and applied to the ML-KNN algorithm.The first suggestion is adding the weight of the sample data of the large sample and the small sample in the mark,so the weight of the small sample is increased by reducing the weight of the data sample Small sample data to reduce the multi-label learning misjudge rate,to achieve uniform sample sampling,effectively reduce the category of imbalance.Second,because the ML-KNN algorithm is a first-order algorithm,without considering the correlation between the mark information,and the correlation between the mark information can be effectively used to improve the learning effect,so the ML-KNN algorithm,Function of the function of a certain change.Finally,the improved ML-KNN algorithm is compared with the existing multi-mark learning algorithm.And theimproved validity is demonstrated.In the part of conclusion,the paper overviews the thesis and puts forward some practical suggestions for the online retail sellers.At the end of this paper,two shortcomings and the three points of prospect are also put forward.