The Research And Implementation Of Demande Prediction And Recognition In Customer Relationship Management

Data mining technologies have offered useful tools of data analyses formarketers in customer relationship management (CRM). With the prosperousdevelopment of e-commerce, CRM is an underexplored filed that has manyopen and interesting problems to the industry and academia. Customer-centricand customer-oriented management has become the key factor in businesssuccess. In order to better maintain and develop the relationship betweenenterprise and customers, in this paper, we built predictive models ofcustomer demand on different levels of customer segmentation; explored theinfluence of interaction between enterprise and customers on the evolution ofcustomer demand; and identified customers’ personalized needs inrecommendation system. Customer segmentation is an important approach inCRM. Depending on the level of disaggregation, marketing activities can bedivided into aggregate, segmentation and1-to-1marketing approach. Thispaper discussed how different data mining methods can be applied in demandforecasting and identification on these different segmentation levels.Sales forecasting on aggregate marketing level is very important forproduction planning of large-scale retail enterprises. Compared to traditionalretail channels, e-commerce deals with a greater volume of transaction and avariety of distribution channels, so more accurate sales forecasts will avoidcustomer churn due to insufficient supply. In this paper, in order to improvethe accuracy of sales forecasts under different marketing environment, weexplored artificial neural network and support vector machine, and made fulluse of the abundant information of customer profile and transaction. Although aggregate marketing is important for forecasting anddecision-making of enterprise, segmentation marketing plays a leading role indeeply understanding customers. However, few rigorous studies exist thatcompare the advantages and disadvantages of these approaches. In this paper,we conducted simulation experiments and compared the predictiveperformance of aggregate, segmentation, and1-to-1marketing approaches.Experimental results show that1-to-1marketing significantly stands out inprediction and recognition of customer demand.Personalized marketing, as the principle idea of learning and managingcustomers, initialized the use of recommender system for personalization inCRM, especially under e-commerce environment. Different from the purerecommender platform, in order to better plan marketing activities, whenproviding personalized recommendation, enterprise need to consider theimpact of interaction between enterprise and customers on customer choices.So in this paper, we fully explored the temporal dynamics in recommendersystem. In order to track the drift of customer preferences over time, and howcustomer choice is influenced by external factors during the interactionsession, we proposed a uniform framework to integrate the long-andshort-term temporal dynamic factors in recommender system. Moreover, thefinal aim of personalized marketing is to increase customers’ lifetime valueand maximize the business profitability, so we also discussed therecommender system that considers the profitability factor for enterprise.