Many scholars have produced a number of definitions of the virtual community. Rheingold was first defined the virtual community in 1993. He thought that the virtual community was fixed personal relationships formed by long-term active participation of community users in community discussion [42H. Rheingold, The virtual community: Homesteading on the electronic frontier., MIT press: London, 1993.]. Since then, many other scholars have conducted corresponding studies and proposed various definitions. Through study of the representative concept [43L. Komito, "The Net as a foraging society: Flexible communities", Inf. Soc., vol. 14, pp. 97-106, 1998.
[http://dx.doi.org/10.1080/019722498128908] -48X. Xiaolong, and W. Fanghua, "Research frontier of virtual community", J. Foreign Econ Manag, pp. 10-16, 2007.], we found that common interests or objectives, internet media and virtual social attachments, or even love are the three forming essential factors of a virtual community.

Fig. (1) Brand community model.

Muniz and O’Guinn have for the first time set comprehensive definitions on brand community: it is a specific, non-geopolitical community based on the complete social relationship between the consumers of the brand, and they put forward a triangular relationship model [49A.M. Muniz Jr, and T.C. O’guinn, "Brand community", J. Consum. Res., vol. 27, pp. 412-432, 2001.
[http://dx.doi.org/10.1086/319618] ]. While McAlexander, Schouten and Koening think that the formation of the brand community was driven by consumers’ common interests in same brand and given out a crucial consumer model [50J.H. McAlexander, J.W. Schouten, and H.F. Koenig, "Building brand community", J. Mark., vol. 66, pp. 38-54, 2002.
[http://dx.doi.org/10.1509/jmkg.66.1.38.18451] ]. Two representative brand community models are shown in Fig. (1).

Virtual brand community is a combination of the virtual community and brand community and has the following characteristics: 1) it is internet-based and not restricted by time and space; 2) it is professional; 3) it is purposefully established: enterprises desire to promote products or maintain their image, consumers desire to share product knowledge or user experiences, or third-parties desire to help consumers gain product information.

Many scholars have also conducted research about the virtual brand community. Kozinets holds that virtual brand community is a community where its members exchange their knowledge or experiences about brand/product on microblogs, BBS or forums [51R.V. Kozinets, "The field behind the screen: using netnography for marketing research in online communities", J. Mark. Res., vol. 39, pp. 61-72, 2002.
[http://dx.doi.org/10.1509/jmkr.39.1.61.18935] ]. From the macro-point of view, Liyin Jin defined an online community as one in which members communicate with each other taking the core of brand is virtual brand community [52"The effects of value dimensions on sense of community, loyalty and brand related behavior intentions in virtual brand community", J Manag Sci, vol. 2, 2007. Available from: http://en.cnki.com.cn/Article_en/CJFDTOTAL-JCJJ200702005.htm]. In this paper, we selected Xiaomi Forum, a community set for expanding famous degree, understanding customers' needs and recognizing the innovation opportunities by Xiaomi Tech, as the object of our study.

This paper studies the user value contribution in a community but does not involve storage and distribution of knowledge in material carriers. As a result, we only focus on two types of knowledge networks in this work:

(1) Knowledge network: Connections among all the knowledge. The model is G_k (K, E_k-k). Thereinto, K = {k₁, k₂,...k_n} denotes point sets of knowledge in the virtual brand community and denotes edge sets. Boolean variable θ (k_i, k_j) = 1 is introduced to represent the co-occurrence of knowledge i and knowledge j in a post.

(2) Member network: Connections among all the community members. The model is G_p = (P, E_p-p). There into, P = {p₁, p₂,..p_n} denotes members in virtual brand community and denotes edge sets. Boolean variable θ (k_i, k_j) = 1 is introduced to represent the interactive relationship between member i and member j in a post.

The relationship between the knowledge network and the member network can be represented as is introduced to represent that knowledge k_j is mastered by member p_i. There are two mapping relationships which could reflect the distribution of knowledge resources in a life carrier or the situation of grasping knowledge of members. The corresponding mapping relationships is as simple as follows:

1. Mapping from knowledge to member: denotes a set of members who has mastered knowledge k_j
2. Mapping from member to knowledge: denotes a set of personal knowledge of member p_i.

Although the knowledge network, the member network and their relationships may help us to define the distribution of knowledge resources in a life carrier as well as find and locate the specific knowledge, they cannot measure the value of knowledge or of a user. However, a weighting process based upon frequency to various relationships is an absolutely fantastic solution to this quantification problem.

1. Connections of the knowledge network are introduced to represent the weight set of E_k-k, which marks the compactness degree of co-occurrence relationship of knowledge.
2. Connections of member networks are introduced to represent the weight set of E_p-p, which marks the compactness degree of user interactive relationships.
3. Connections between knowledge and a member are introduced to represent the weight set of (E_p-k) which marks the mastery degree of knowledge k_j by user p_i.
4. Knowledge stocks: Q(k) = {q(k_j)} are introduced to represent the weight set of knowledge storage q(k_j) denotes storage of knowledge k_j of all users.

According to user-contributed content value in the virtual brand community, we can find and locate key users who grasp important or scarce knowledge, then scoop out the deeper customer need and encourage users to participate in product design, development and improvement to finally realize innovation. This means that our objective is finding important information and users who have mastered this information. This objective could be achieved through a WKSN-based user-contributed content value model, which takes users and knowledge as nodes and user interaction and co-occurrence of knowledge as the relationship between nodes, as shown here:

The construction method is shown below:

1. Obtain point sets of users and knowledge
   This paper applies data mining technology in data collection to obtain user information and knowledge content. All users constitute the point set of users. Point sets of knowledge are obtained by text mining whose core technologies are preprocessing, segmentation and extracting feature words [68T. Chen, and Y. Xie, "Literature review of feature dimension reduction in text categorization", J. China Society Scin. Tech. Informn, vol. 24, pp. 691-695, 2005.], and combined with ontology and domain knowledge dictionary, etc.
2. Determine the relationship set
   User relationship set (E_k-k): the interactive relationship (post or reply) between members in a post is the edge of the user network.
   Knowledge relationship set (E_p-p): the co-occurrence relationship between knowledges in a post is the edge of the knowledge network. Set of relationship between knowledge and user (E_p-k): the connection between knowledge extracted from posts and users publishing this post is the edge of the network.
3. Calculate the weight set of knowledge stock of all users
   We apply the frequency of knowledge in all users’ knowledge set as Knowledge stocks and constitutes the weight sets of knowledge stock of all users.
4. Calculate weight set: W (E_k-k), Q (E_p-k)
   W (E_k-k) represents the concurrent times of knowledge in different posts. The more times, the bigger knowledge relevance and higher the weight. Q (E_p-k) is measured by the absolute value of frequency of knowledge in all posts by each user.
5. Build the model

Based on the above computation and combined with research findings, we can construct the WKSN-based user-contributed content value model.

People who registered an account are members in the virtual brand community, and those members’ interactions (post or reply) form weakly tied social networks. Some prominent users are likely to hold the positions of structural hole and foster and support the free flow of useful information. However, potential negative effects generated by them should not be underestimated. For the perspective of enterprises, managers need to grasp user relationship information to find users in the positions of the structural hole and lead them to spread beneficial information. Moreover, this objective could be achieved through a Weighted Social Network Analysis-based user interaction value model, which takes users as nodes and user interactions in same post as the relationship between nodes, as shown here:

The construction method is shown below:

1. Get point set of users
   • All users constitute the point set of users.
2. Determine the user relationship set: E_p-p
   Interactive relationship (post or reply) between members in a post is the edge of network.
3. Calculate weight set: W (E_p-p)
4. Build the model
   Based on the above computation and Combined with research findings, we can construct WSNA-based user interaction value model.

First and foremost, we analyze the weighted knowledge network, which takes knowledge as nodes and co-occurrence of knowledge as edges to find core knowledge. As shown in Fig. (3).

Fig. (3) Weighted knowledge network in virtual brand community.

Fig (4) Centrality analysis of weighted keyword network.

Fig. (5) Core keyword frequency statistic.

It is clearly observed that most instances of knowledge are connected together, while a small percentage of keywords, such as words and licenses, are independent of integrity. We can find in a preliminary estimate the core knowledge or focus area of users. Furthermore, we adopt a centrality index [69A. Bavelas, "Communication patterns in task-oriented groups", J. Acoust. Soc. Am., vol. 22, p. 725, 1950.
[http://dx.doi.org/10.1121/1.1906679] , 70G. Sabidussi, "The centrality of a graph", Psychometrika, vol. 31, no. 4, pp. 581-603, 1966.
[http://dx.doi.org/10.1007/BF02289527] [PMID: 5232444] ] to measure the “power” of keywords in the network in order to analyze the current status and characteristics of users’ knowledge deeply. Operation results are shown in Fig. (4).

Based on the Pareto Principle, this paper sort keywords according to three centrality indices and counts the word frequency of top-ranking keywords. Finally, determining top 20 keywords to a total word frequency of 5094 as core knowledge. The results are shown in Fig. (5).

Fig. (6) Weighted core keyword network.

Besides, draw weighted top-keywords network which takes top 20 keywords as nodes, co-occurrence of knowledge as edges and co-occurrence frequency as edge weight. As shown in Fig. (6), we can see the compactness degree of relationship among keywords.

Fig. (7) WKSN-based user-contributed content value model.

Last but not least, based on above-mentioned analysis and construction method of the user-contributed content value model, we can draw a weighted knowledge super-network which is mainly composed of the top 20 keywords network and corresponding user network. Specifically, empty circles represent keywords, filled circles represent users, edges represent interaction relationships, co-occurrence relationships, or mapping relationships, as shown in Fig. (7).

Fig (8) Evaluation results of user-contributed content value.

As you can see from Fig. (7), the model is able to reflect the distribution and stock of each instance of knowledge in the community, and it also can reflect the knowledge mastery condition of each user.

This paper applies the numerical magnitude of core knowledge storage to evaluate the user-contributed content value. A higher value number corresponds with a greater contribution of the user. As a consequence, the operational result of user-contributed content value by Pajek is shown in Fig. (8).

Fig. (9) Weighted user network of WSNA-based user interaction value model.

According to the operational results, hooyeah is the largest contributor to community knowledge and far exceeds other users. ErCiChuangYe ranked 2^nd with 409 points. Although MiFenLiRen as well as other users have taken part in community interactions, their contribution value is low.

We can draw the weighted user network, which takes 9230 users as nodes, interaction relationships as edges and number of interactions in posts as weight of edges by Pajek. The results are shown in Fig. (9).

Fig. (10) Calculational results of degree centrality.

User interaction network is based on weak ties in the virtual brand community. Moreover, the research objective here is to find influential and prestigious user in community. Therefore, we can use social network analysis technology to evaluate user interaction value.

A centrality index [71L. Jun, Lectures on Whole Network Approach., vol. 15. Shanghai People's Publishing House: Shanghai, 2009.] of social network analysis technology is a useful tool for researcher to find core members in a group. It mainly contains three indices: degree centrality, closeness centrality, betweenness centrality.

Fig. (11) Calculational results of betweenness centrality.

1. Degree Centrality

Degree centrality measures the number of nodes connected to a member. The larger the number is, the more active the member is. The active members have more influence than other less active members. From the operation results Fig. (10), there are great differences among nodes, with hooyeah having the largest degree centrality, which means his position is of vital significance.

Fig. (12) Calculational results of closeness centrality.

User interaction value denotes the users’ ‘power’ in the social network, characterized by the influence, control force and dissemination capacity. We think users contribute to the online word of mouth exchanges when and only when his/her three indices are high. According to generated result of user centricity, we sorted the users in the community and the results are shown in Table 2.

From Table 2, we find that hooyeah, Jiao WoHei Zhu and ErCi Chuang Ye and other users occupy an important role in the social network’s structure because their degree centrality, closeness centrality and betweenness centrality are at the top of the corresponding list.

Table 2

Centricity sequence in WSNA-based user interaction value model.

Based on the above analysis of user value contribution in the virtual brand community, users can be sorted by user-contributed content value and user interaction value and the corresponding evaluation indices as follows:

(1) Evaluation index of user-contributed content value.

In this paper, user-contributed contents are simplified and measured by keywords of posts and user-contributed content value was evaluated by core knowledge storage of users.

(2) Evaluation index of user interaction value.

This paper chooses degree centrality, closeness centrality and betweenness centrality as the evaluation index of user interaction value.

In this paper, users are grouped through correlation cluster analysis based on two evaluation indices of user-contribute value, and are classified into four relatively independent categories. The method we adopt is k-Means proposed by McQueen in 1967. This method clusters items into k clusters based on the distance the items are from the centroid of the previous iteration. Specifically, each user is uniquely identified with a username and four variables (user-contributed content value, degree centrality, closeness centrality and betweenness centrality) are used in user classification by using the Q-type cluster. The results of the clustering analysis are shown in Tables 3-6:

Table 3

Initial clustering center.

Table 4

Final clustering center.

Table 5

The distance between the final clustering center.

Table 6

The final case number in each cluster.

With the results of clustering, users in the community can be divided into four categories (Table 7):

Table 7

User segmentation.

From the ANOVA Table 8, it can be seen that the four variables have significant positive correlation with user classification. This illustrates the rationality of the evaluation index and the feasibility of the Q-type cluster.

Table 8

ANOVA table.

User-contributed content value and user interaction value can be used to construct the user classification matrix which is shown as Fig. (13).

Valuable users: while the proportion of these users is lowest, they add the most benefit to the virtual brand community. They are willing to contribute their own knowledge and actively participate in community interaction. Therefore, valuable users deserve serious attention from enterprises.

Knowledgeable users: these users have and love to share rich user experience and unique insight of products. Nevertheless, they have weaker connections with others in the community. Albeit the number of these users is relatively small, they are professional. As a consequence, knowledgeable users are important messengers of users’ needs for community managers.

Social users: these users take an active part in community interaction with great influence, in spite of the fact that their knowledge contribution is relatively low. As a result, social users play an important role in marketing and information propagation.

Fig. (13) User classification matrix in the virtual brand community.

Regular users: most of the users in the virtual brand community are regular users. They rarely interact with others and share experiences but depend on the community to a certain degree for useful information. Enterprises should take actions to tie-in users because they are on the edges of the community.

The objective of valuable user management is maintaining a stable relationship between enterprise and users. Enterprises can formulate management strategies from two perspectives: on one hand, giving a material and spiritual reward. Materially, utilizing community manager even enterprise staff when he/she makes a specific contribution or give him/her a special purchase funnel with a certain discount. Mentally, making user’s contribution clearly by user rank. On the other hand, developing a one-on-one or even management strategy to find core knowledge and creative point shared by valuable users so that enterprise can better satisfy the demands of consumers and improve the product competition in the markets.

The objective of knowledge user management is encouraging users to further share their tacit knowledge and propose innovation spots for innovation. We can find that knowledgeable users are mainly concentrated in general groups which is more professional in the community. All members hoping to enter the group must pass the audit of community manager. As a consequence, in order to un-upgraded users, community manager should evaluate user contributions regularly and update users’ groups timely without application and audit to give them the approval of the society and enhance their enthusiasm. To upgraded users, community managers should assure the safety of users’ personal information and improve material incentives.

The objective of social user management is giving full play to their prestige and power in interaction networks. Community managers can formulate management strategies from three directions. To begin with, allowing social users to organize promotion activities online or offline to improve their influence and self-gratification. In addition, giving them certain financial incentives including price concessions and strengthening their participation consciousness according to their activity significantly in the community. Finally, guiding these users to study product knowledge and functions in corresponding sections of their own initiative. Keeping social users’ sustaining influence through interacting with others with the view of specialization.

Regular users are the principal composition of the virtual brand community. They expect professional knowledge in the community, privilege of purchase and so on and so forth. The objective of regular user management is preventing user loss and facilitating translating user to valuable users, knowledgeable users or social users. The community should create a fair, friendly, advanced and reliable atmosphere to improve the community identity of users. In addition, optimizing community capabilities and adding something of interest to increase users’ desire of participation.