Consumer Segmentation And Recommendation Applied Research Based On Web Mining

With the development of information technology,especially with the rapid development of e-commerce,the information volume continues to increase,so we need an effective analytical tool to analyze the daily amount of information generated. After a decade the development of the core technology of data mining has made great achievements.To e-commerce enterprises,from different aspects to establish good relations with customers,improve customer loyalty,without doubt can bring the huge profit in various aspects for the enterprise.The individuation of e-commerce breeding lives,and In the individuation of the content covered,the recommendation service for the customer is the most important,because it can identify different customers,and provide users with information to meet the demand.E-commerce recommendation system can effectively retain customers,prevent customers from running off,improve e-commerce sales volume and enhance the competitiveness of e-commerce website,and the individuation recommendation deepeneds this influence:So e-commerce recommendation system in e-commerce has good prospects for development and application,this article has made useful explorations andresearch for e-commerce recommendation system,in the recommended system,the rules are usually generated off-line,the recommended in accordance with the rules is done by recommended algorithm。Recommended technology can be used based on association rules,based on clustering,based on nearest neighbor technology,based on project grading forecast algorithms and so on coordination filtration,。But along with the information and visit user exponential order's increase,these recommendation algorithm appeared too huge and complex, and seriously affect the efficiency,resulting in a lot of websites do not use these recommendation technology.In view of this question,this article first proposed carries on the classification of the goal customer,also is the consumer subdivides,then makes the different recommendations again to the different consumers,realizing the individuation recommendation.The definite content is as follows:The consumer subdivides,the consumer subdivides is the process which has the heterogeneous characteristic consumer to carry on the cluster,each kind of consumer which after subdivides has the similar demand and the purchase characteristic.In the website,consumer's purchase demand and the purchase characteristic are reflected when compares the choice commodity,also is the browsing commodity behavior of the consumer in the website,we may carry on the WEB excavation through the website browsing diary.The WEB excavation generally is based on the analysis of all website pages。Because this article must carries on the analysis to consumer's purchase demand and the purchase characteristic,therefore only retained the final product information page in the data pretreatment stage,removes other transition,convergence pages,for example products primary categories page,search page and so on.In this paper,starting from the product to find out the consumer full identify path to browse products,transforms product group which glances over for the consumers。But because of different consumer preferences and website design and other reasons,possibly the product which likes regarding the identical group can have the different browsing order,therefore here we will possess the same product the way to unify into a disorder product group.Then carries on the cluster analysis through these disorderly product group。According to Song et al's WEB excavation algorithms, directly carries on processing to the Web stand's topology and the user browsing information,namely take Web stand URL as the line,take UserID as the row, establishes URL - UserID relation matrixs,the element value is user's visits.Based on this,carries on the analysis to a row vector to obtain the similar customer community, carries on the analysis to the course quantity to be possible to obtain the related page. But this article only needs to carry on the analysis to the course quantity,we obtain the product cluster through the procedure.First,we must carry on the quantification to the commodity.Here we divide into five aspects to analyze,respectively is:Price,Brand,Top carriage time,Function, Outward appearance design.Then the comparative analysis product's cluster,has obtained the digital commodity website consumers' characteristic which this article studies.According to the above analysis,we need to withdraw the different rule to cover the cluster result,we discover the cluster of majority products,also is the consumer glances over the product group price difference not to be big,the standard deviation is smaller than 300,therefore we formulate the first rule,the price difference is smaller than 300 converges the price sensitively;Next has quite partial product group's top carriage time is away from the browsing time less than 6 months,also is opposite in other product,was new product,we formulated the second rule,the standard deviation was bigger than 300,and the top carriage time was smaller than six month-long was the happy new kind;Finally also some kind of product group is belongs to the identical brand,may also understand that has own characteristic for each brand,regardless of being or is in externally the function,has their explicit localization,but these are exactly,therefore has someconsumers who the consumer likes obvious by chance and a loyal brand,we formulate a final rule are label the difference to be bigger than 300,and converges the brand for the identical brand to like.Comes through these three rules the consumer to divide into three broad headings:Price sensitive,happy new,brand hobby.And compiles the related SQL language the product group which glances over to all consumers to carry on the confirmation,finally indicated that the rule is correct,we may subdivide the consumer into three kinds.Finally we according to consumer's subdividing,carries on the recognition and the judgment to the online browsing's consumer,forecast that consumer's demand characteristic,makes the recommendation.Compared to the case of the former recommendation algorithm,this step is easy,because we,so long as records the consumer to glance over the product page,according to the front regular order and the content,first calculates all the current browsing product's standard deviation of price whether to be bigger than 300,if also in its acceptable price recommends the products paying biggest attention to in the price sensitive category,also is the standard deviation of recommendation product is not bigger than the standard deviation of the consumer original browsing product group.When the price standard deviation of the product which the consumer glances over is bigger than 300,then we continue to carry on the classification of the consumers by the second rule,also is judging the happy new type,judges the computation so repeatedly.By our statistics before,the rule should be able to cover more than 90%consumers.Besides recommendation application,because had already carried on the segmentation to the consumer,and obtains the characteristic of the consumer purchases,we may also carry on the improvement of the stand,let the consumer in browsing time is more convenientand more direct to choose their own needs,initiative gives oneself to locate the price,the localization brand classification,attention to new products and so on;Even we may also carry on some pointed marketing strategy, supplies the preferential benefit information to the consumer who browse product group price standard deviation specially little,bestows the brand souvenir to the brand hobby consumer and so on to be able to attract the purchase promotion.In summary,the method based on WEB mining consumer subsection and online consumer identify proposed by this paper,in the general case,has a certain theory meaning and application value.However,there are still deficiency in recommended methods,can not contain fault-tolerance function and so on,and for the first time click on the product can not accurately judge the classification of consumers,these problems have to be studied further.