Abstract: [Purpose/Significance] Digital communities have become a way for enterprises to manage users efficiently. The existing research on digital community rarely considers the importance of user behavior information and user's customer life cycle value to the mining of user preferences in digital community. This research aims to give full play to the digital community's characteristics such as intuitive, convenient, interesting, and interactive properties so that the research results can benefit every user in their use of the digital community and every enterprise in their user management. [Method/Process] Aiming at the user groups in digital community, this paper proposes a preference mining model ClV-Preference mining (CLV-PM) based on Customer Lifetime Value (CLV). First, in order to reflect the real preferences of users, the three indicators of the RFM model are used to quantify user behavior information, and the group characteristics of users are mined through K-mean ++ algorithm to generate user value category labels. Second, in order to consider the timeliness and difference of users and enhance the model's cognition of preferences, this paper uses the entropy weight method to solve the indicator weights of each activity, obtains user CLV to construct user-project scoring matrix, and uses the collaborative filtering algorithm to predict user preferences. Finally, based on the user value category, user historical preference and user forecast preference, the user preference profile of target users in digital community is generated, and feasible suggestions are put forward for the operation and maintenance of target users according to the user preference profile. [Results/Conclusions] The user dataset of the "Wechat community" management platform can be divided into four user value categories: important value users, low value users, returned users and important retention users. Target users 16254 are important value users, and the operation strategy of "retention and maintenance" is adopted. The historical preferences are happy hop, sec-kill and other activities; the prediction preference is flying chess battle, guessing code map and other activities; the target user preference sketch provides the basis for the operation and maintenance of users in the digital community. In terms of data source, the CLV-PM model proposed in this paper directly reflects user preferences based on user behavior information and reduces the problem of score distortion. To provide a new perspective for the research of user behavior in digital community, the construction of user-project scoring matrix based on user CLV fully considers the user value of digital community and provides a new direction for the extension and application of CLV. However, due to limited research space, this paper did not conduct model evaluation research on the proposed model, which can be further discussed in subsequent studies.

Key words: CLV-PM, collaborative filtering, digital community, user preference, information behavior