Build a customer identification model.

The established customer preference identification model uses data mining technology to collect, sort out, and analyze the basic information and online behavior data of customers on the we-media platform so as to identify and lock customers according to consumption characteristics and other indicators. Construction of individual customer identification database: when building a customer identification model, the first step is to mine the behavior characteristic data of individual customers (original target) to form the basic label, that is, to build the individual customer identification database. Through the study and sorting out of the previous research studies of scholars, the “consumer graph” constructed in precision marketing based on the five dimensions proposed in the portrait of social media with high conversion rate is studied. Based on the “4C” theory and combined with the data features used, the framework of individual customer identification database is proposed, which mainly includes two aspects: basic customer information and online behavior data, as shown in Figure 2.

Community research is carried out on the basis of the establishment of individual customer identification database. This step makes the customer identification model more comprehensive and radiates its friends, family members, and colleagues through individual customers. Each individual customer as the center of the circle to form a community, the intersection of multiple communities will be able to make the individual more comprehensive label.

Cross-sectional research is to lock and mine social data on the we-media platform of people related to the original goal (such as individuals with frequent interactions and users with clear groups), enrich the basic label of the original target, and describe it from different latitudes and granularities. With the popularity of social software, individuals have diversified their use of social software. In order to distinguish work and life, they will apply for multiple accounts on the same social software; therefore, longitudinal research is to mine the data of different social accounts of the original target, including different accounts on the same platform and accounts on different platforms (by default, an individual has only two accounts at most on different we-media platforms). Comprehensive identification can be performed based on the login device or login IP address. The social network analysis method is introduced above, which is a quantitative method to formally describe the social network. A social network consists of a collection of nodes and wires, each node representing an individual (the original goal), connections indicate relationships, which can be relatives, friends, colleagues, or even strangers with the same commodity preferences. Individuals are connected through relationships and eventually form social networks, each label sharing mutual reference, then, form each three-dimensional user. The original target is at the heart of the social network, users who are more connected and more active online than other customers, the larger and more complex the social network, the more comprehensive the customer information will be. The social network structure is shown in Figure 3.

Identify and analyze preferences.

Based on the customer identification model, data related to consumption preferences are mined to identify and analyze consumption preferences. The steps and methods are as follows: first, consumer interest mining based on online social data was performed. For the discovery of customer interest, in the process of online customer participation in information dissemination, different interests and concerns will be presented at different times, which can be called dynamic interest migration. Firstly, the characteristics of feature items as customer interests in the dynamic migration process are analyzed, and the text feature extraction method is proposed. On this basis, the finite mixed probability model is used to study the identification method of customer interest and the merging method of new interest content is also used. The network customer is the main body of information transmission, and the customer with large centrality plays a vital role in information transmission. The preliminary research found that in online social network, the customer nodes with a large degree of connection can help to spread information to more customers, and the customer nodes with a large number of medium can help to expand the range of information transmission, customer nodes with large aggregation are helpful to local information propagation, and the positive and negative attributes of node connection can influence the choice of information propagation path. Based on these structural attributes, important customer nodes with different centrality can be selected so as to provide important customer attributes for the establishment of specific types of customer behavior models. Second, customer consumption behavior pattern mining based on online social data was performed. In online social networks, the behavior patterns of some subsets of customer nodes are similar to each other due to the similar roles, concerns and interests of certain types of customer nodes. Therefore, on the basis of large-scale customer node behavior data, based on the convergence of customer node behavior patterns and the similarity of behavior characteristics, the antinoise clustering algorithm is used to analyze the complete set of customer nodes, and several subsets of customer nodes with similar behavior patterns are obtained. Furthermore, the frequent item association rule mining algorithm is selected to analyze a subset of customer nodes, and the main behavior patterns of customers in the subset can be obtained. Thirdly, customer preference recognition and analysis based on customer consumption behavior pattern was performed. Through semantic analysis, customer interest and behavior data obtained from online social media and consumer behavior tendency data obtained from other types of data sources are analyzed. The change and development process of different types of consumption behaviors was recorded in chronological order and formed a time series. Based on spatial reconstruction, the measured values were constructed into a group of multidimensional dynamic vectors [12]. The large difference in the time series means that the consumption behavior tendency contained in the original data changes greatly. This analysis is therefore able to identify shifts in consumer behavior, and due to the introduction of new types of data, such as online social networks, some changes that cannot be reflected in the consumer behavior data released based on the traditional questionnaire survey can also be identified so as to make the identification results more accurate and perfect.

In order to implement the prediction of new commodity buyer behavior based on decision tree induction, this chapter uses the sales transaction data of Jingdong Mall. Taking into account the changes in the initial form of Jingdong Mall, this paper only selects the newly listed products in memory category in 2009 and 2010. In order to avoid the impact of the website’s anniversary promotion on customers’ purchasing behavior, the paper finally selects the newly listed products from July 2009 to February 2010. After the new product is put on the shelves, the personal page links of the top 20 customers who bought the product first will be obtained according to the purchase time of the customers. Then, the personal purchase history will be entered into the personal purchase history to obtain all the purchase history of the new product on the website. The training set was constructed according to the characteristics obtained by the above method, and the data of customer purchasing behavior was predicted, as shown in Table 1.