Contents lists available at SciVerse ScienceDirect
Electronic Commerce Research and Applications
journal homepage: /locate/ecra
目录列出了可用的华艺SCIENCEDIRECT
电子商务研究与应用
杂志主页：/定位/ ECRA
Discovering target groups in social networking sites: An effective method for maximizing joint influential power
在社交网站中发现目标群体：一种有效的方法为最大限度地发挥联合的影响力
With the tremendous popularity of social networking sites in this era of Web 2.0, increasingly more usersare contributing their comments and opinions about products, people, organizations, and many otherentities. These online comments often have direct influence on consumers’ buying decisions and the pub-lic’s impressions of enterprises. As a result, enterprises have begun to explore the feasibility of usingsocial networking sites as platforms to conduct targeted marking and enterprise reputation managementfor e-commerce and e-business. As indicated from recent marketing research, the joint influential powerof a sm all group of active users could have considerable impact on a large number of consumers’ buyingdecisions and the public’s perception of the capabilities of enterprises. This paper illustrates a novelmethod that can effectively discover the most influential users from social networking sites (SNS). In par-ticular, the general method of mining the influence network from SNS and the computational models ofmathematical programming for discovering the user groups with max joint influential power are pro-posed. The empirical evaluation with real data extracted from social networking sites shows that the pro-posed method can effectively identify the most influential groups when compared to the benchmarkmethods. This study opens the door to effectively conducting targeted marketing and enterprise reputa-tion management on social networking sites.
随着广受欢迎的社交网站在这个Web 2.0时代，越来越多的用户有关产品，人员，组织，和许多其他贡献自己的意见和建议的实体。
这些网上的意见往往直接影响消费者的购买决策和酒吧LIC的企业的印象。
其结果是，企业已经开始探索使用的可行性社交网站平台，进行有针对性的标志和企业声誉管理电子商务和电子商务。
正如从最近的市场调研，联合影响力一小群的活跃用户，可以有相当大的影响消费者的购买大量的决策和公众的认知能力的企业。
本文阐述了一种新型的方法，可以有效地发现最有影响力的社交网站（SNS）的用户。
在PAR-特别地，开采的影响从SNS的网络的一般方法，和计算模型发现的用户群体最大的联合影响力的数学规划是亲构成的。
从社交网站中提取的真实数据的实证评价表明，亲构成的方法可以有效地识别最有影响力的群体相比基准的方法。
这项研究打开大门，有效地进行有针对性的营销和企业声誉在社交网站上的重刑管理。
随着Web 2.0时代的到来，社交类网站越来越受到欢迎，越来越多的产品，组织，人员，和许多其他用户把自己的意见和建议实施和添加到这个实体中。
这些社交类网站上的意见往往直接影响消费者的购买决策和对于LIC的企业的印象。
其导致的结果，就是企业已经开始探索开发和使用自己的可行性社交网站平台，在电子商务中进行有针对性的和对于企业声誉有关的管理。
最近的市场调研显示，联合一小群有影响力的活跃用户，可以在相当大的程度上影响消费者的购买决策，文阐述了一种新型的方法，可以有效地发现最有影响力的社交网站（SNS）的用户。
特别的，从SNS的网络提起的有一定影响的开发方法，与利用计算机模型发现的最大的且具有强力影响力的用户群体有相互关系。
从社交网站中提取的真实且具有实证评价的数据表明，亲构成的方法可以有效地识别最有影响力的群体相比的方法。
这项研究有效地打开了有针对性的营销和在社交网站上的管理企业声誉的大门。
1. Introduction
1。
介绍
In this Web 2.0 era, users are able to express their opinions onproducts through many channels such as online forums, shoppingwebsites, blogs, and wikis. These opinions can influence otherusers’ buying decisions and their views on companies (Cheunget al. 2009, Chevalier andMayzlin 2006, Dellarocas 2003,Hennig-Thurau and Walsh 2003, Koh et al. 2010, Mayzlin 2006,Park et al. 2007). Recently, the emerging channel of social network-ing sites (SNS), such asFacebook, Twitter, and Epinions, has at-tracted the attention of marketing practitioners and researchers.These sites not only permit users to express comments and opin-ions on products, people, organizations, and many other entities,but also enables users to build various social relationships. Forexample, on the Epinions site (Epinions 2010), a user can build atrust relationship with another by adding him or her to a trust list,or the user can block him or her with a block list. The site thenshows the trusted users’ opinions at the top of the list. With thesesocial relationships,
在这个Web2.0时代，用户可以表达自己的意见产品通过多种渠道，如网上论坛，购物网站，博客和维基。
这些意见可以影响其他用户的购买决策，他们的意见对公司（长安等人。
2009年，富安，2003年Dellarocas2006年Mayzlin的Hennig图劳和沃尔什2003年，苏梅等。
20102006年Mayzlin，Park等人。
2007年）。
最近，新兴的社交网络渠道（SNS）网站，如Facebook，Twitter和Epinions-牙牙营销从业者和研究者的关注。
这些网站不仅允许用户表达意见和奥平离子对产品，人员，组织，和许多其他实体，但也使用户能够建立各种社会关系。
为的的Epinions网站（Epinions2010）上，例如，用户可以建立一个信任关系与另一个他或她加入到信任列表，或者用户可以阻止他或她的阻止列表。
然后该网站显示在列表顶部的受信任的用户的意见。
与这些社会关系，
pinions will have greater impact on usersthan those expressed on the other channels (such as shoppingwebsites) because people always believe or accept more easilythe opinions of those with whom they have social relationships(Golbeck 2005, Lu et al. 2010, Massa and Avesani 2007). In addi-tion, the influence of opinions on SNS can be disseminated morewidely and quickly than that of other channels. Thus, some users’opinions captured on SNS can greatly influence other users’ buyingdecisions or their views on certain companies.
Many business entities have recently come to recognize thisphenomenon, and some companies have begun to identify certainusers of SNS to conduct online marketing and reputation manage-ment (Conlin and MacMillan 2010, Marks 2010, Miller and Dickson2001) in e-commerce and e-business. For companies to better uti-lize SNS for cost-effective, targeted marketing and reputation man-agement, they must address an important question, given the hugenumber of social network users a nd companies’ limited budgets.That brings us to the question of which users’ opinions will most influence others’ actions. If the most influential group of userscould be identified, companies could consume minimal resourcesto improve product sales and enhance their reputations.
齿轮将有更大的影响所表达usersthan对其他渠道（如shoppingwebsites）的，因为人们总是相信或接受更easilythe的那些与他们有社会关系（Golbeck2005年，鲁等人，2010年，马萨和2007年Avesani的意见）。
此外，SNS 上的意见，能散发morewidely和速度比其他渠道影响。
因此，一些SNS上拍摄的users'opinions可以大大影响其他用户的buyingdecisions对某些公司或他们的意见。
最近许多商业实体已经认识到寻找这个状况，一些公司已开始物色SNS certainusers的，在电子商务和电子商务开展网络营销和声誉管理（康林和麦克米伦2010年，2010年商标，米勒和Dickson2001的）。
对于符合成本效益的，有针对性的营销和声誉的人管理公司，以更好地利用丽泽SNS，他们必须解决的一个重要问题，鉴于社交网络用户和公司的有限budgets.That hugenumber给我们带来了哪些用户的问题“意见mostinfluence他人的行为。
如果最有影响力的一群userscould被确定，企业可以消耗最少resourcesto的提高产品销售和提高自己的声誉。
Although there are many existing studies on measuring nodeimportance in social network analysis (Wasserman and Faust1994), as well as studies that explore the spread of influence in so- cial networks (Kempe et al. 2003, Kempe et al. 2005), these works emphasize the importance of each node, without considering thejoint influ ential power of a group of nodes. According to the latestfindings from marketing research (Katona et al. 2007, 2011), if thecustomers are provided with positive information on products orenterprises by all related users in online communities, there maybe a higher probability of customers purchasing such products orhaving positive perceptions of these enterprises. This is known asthe joint influential power of a group of users. Previous researchalso indicates that the joint influential power of a small group of users could have considerable impact on a large number of con-sumers (Domingos and Richardson 2001, Richardson and Domin-gos 2002). Therefore, marketing personnel should identify theusers who have great joint influential power on SNS, and find waysto encoura ge these users to express positive opinions about com-panies and the companies’ products through the strategy oftargeted marketing. As a result, companies could maximally pro-mote product sales and improve enterprises’ reputations throughthe joint influential power of the specific group of users.
Effectively discovering the group of users with maximal jointinfluential power from the huge number of users on SNS has be-come a key issue for companies to conduct targeted marketingand reputation management. Although previous research hasexamined the problem of discovering a group of influential users,the heuristic method used does not address the issue of identifyinga group with maximal joint influential power (Zhang et al. 2008).One of the weaknesses of the previous research is that the usersare added into the target group one by one, according to theirattributes and parts of trust relationships between them, withoutconsidering the influential power of the target group as a whole.Thus, these method usually do not disc over target groups havingthe most joint influential power (Kempe et al. 2003). In contrast,our proposed method represents global influential relationshipsamong all users as a directed graph, and uses mathematical pro-gramming as the computational apparatus to discover the groupof users with maximal joint influential power. Considering the costof marketing, the proposed models can discover the target groupwith flexible costs. The empirical evaluation with real data fromEpinions and Twitter websites shows that the proposed methodshows much better performances, compared to the benchmarkmethods. In summary, the main contribution of our research isthe development of a novel method that discovers the user groupwith maximal joint influential power in a cost-effective way; italso overcomes the disadvantage of existing methods (Zhanget al. 2008) which only consider the attributes of users and partsof influence relationship. Our research opens the door to applywidely available data captured on SNS to conduct targeted market-ing and enterprise reputation management in e-commerce ande-business.
虽然有许多现有的上测量nodeimportance研究在社交网络分析（Wasserman和Faust1994），以及研究，探索中的传播影响力，使的CIAL网络（肯普等人，2003年，肯普等，2005年），这些作品强调每个节点的重要性，而不考虑的影响力thejoint的一组节点。
据从市场调研latestfindings（卡托纳等，2007年，2011年），，如果thecustomers提供所有相关用户在网上社区积极信息产品orenterprises的，有可能是概率较高的客户购买此类产品orhaving这些积极的看法企业。
这被称为的联合asthe影响力的一组用户。
上一页researchalso表示的联合影响力的一小群ofusers的可能有相当大的影响了一大批CON消费者（2001年多明戈斯和理查德森，理查德森和海德明-GOS 2002）。
因此，营销人员应确定有很大的影响力联合对SNS的用户是想怎么样，找到waysto鼓励这些用户表达正面的意见，对公司和公司的产品通过战略oftargeted营销。
因此，企业可以最大限度地促进产品销售和提高企业的声誉throughthe特定的用户群的
联合影响力。
有效地发现了一批的用户最大jointinfluential功率从SNS上的用户数量庞大的已来到进行有针对性marketingand的声誉管理公司的一个关键问题。
虽然以前的研究第十一条的问题，发现了一批有影响力的用户，使用启发式方法不解决这个问题，identifyinga组联合有影响力的最大功率（Zhang等，2008）。
在前人研究的弱点之一是，添加到目标群体之一usersare一，根据theirattributes和部分之间的信任关系，withoutconsidering作为一个whole.Thus目标群体的影响力，这些方法通常没有发现目标群体havingthe最共同的影响力（肯普等人，2003）。
相反，我们提出的方法表示全球有影响力relationshipsamong的所有用户作为一个有向图，并使用数学计算设备发现groupof的联合影响力最大的用户编程。
考虑costof营销，提出的模型可以发现目标groupwith的灵活的成本。
实证评价与真正的数据fromEpinions和Twitter的网站显示，建议methodshows的更好的表演，到benchmarkmethods相比。
综上所述，我们的研究“是一种新的方法，发现用户groupwith最大的联合影响力的具有成本效益的方式发展的主要贡献，它也克服了现有方法的缺点（人Zhanget 2008）只考虑属性用户和影响关系partsof。
我们的研究开启了大门到applywidely SNS上捕获的数据进行有针对性的市场和企业声誉管理在电子商务安山岩业务。
The rest of the paper is organized as follows. Section 2 discussesrelated research. Section 3 provides an overview of the proposedmethod for cost-effective targeted marketing and enterprise repu-tation management. Section 4 presents the computational modelsfor discovering the most influential group s of users. Section 5 re-ports the experimental evaluation of the proposed method, andSection 6 summarizes our research work and discusses directionsfor future work.
剩下的纸张安排如下：第2 discussesrelated研究。
第3节提供了一个符合成本效益的有针对性的市场营销和企业声誉的塔季翁管理概述proposedmethod。
第4节提出了的计算modelsfor发现最有影响力的团体用户。
第5端口的实验评估所提出的方法，andSection6总结了我们的研究工作，并讨论了今后的工作directionsfor。
2. Related work
2。
相关工作
2.1. Target group
2.1。
目标客户群
A few studies exist that involve discovering a target group, and these bear some similarity to this study. In (Zhang et al. 2008), an algorithm based on the trust relationships between users is pro-posed to detect the influential target groups. The users are ranked according to the number of their written reviews, and then the users and the trust relationships are sequentially added into the target group until the clustering coefficient of the target group is less than a threshold. However, the clustering coefficient only re-flects the influence relationships of the users added into the group,without considering the influence of relationships outside the group. Thus, this method fails to consider the influential power of the target group as a whole, and it tends to discover target groups consisting of mutual influence users which do not always have the most joint influential power over other users (Kempe et al. 2003). Diverging from their method, our proposed method mines the influence relationships from SNS, represents them in a directed graph, and discovers the target group that has maximal joint influential power by exploring all influence relationships using the models of mathematical programming.
存在一些研究，包括发现目标群体，而这些有某些相似之处这项研究。
（张等，2008），提出了一种基于用户之间的信任关系是亲对检测有影响力的目标群体。
用户排名按照其书面评论的数量，然后依次加入到目标群体的目标群体是直到聚类系数小于阈值用户和信任关系。
然而，聚类系数只有重新反映用户添加到该组的影响关系，而不考虑的影响外组的关系。
因此，这种方法没有考虑到目标群体的影响力作为一个整体，它往往发现目标群体包括相互影响的用户并不总是有最联合的影响力比其他用户（肯普等，2003年）。
发散他们的方法，我们提出的方法地雷的影响从SNS的关系，表示他们有向图中，发现目标群体，具有极大的影响力联合探索所有影响关系，利用数学规划模型。
2.2. Node importance
In social network analysis, centrality is proposed to measure the importance of one node, and the nodes with higher centrality will then influence others more greatly (Wasserman and Faust 1994). Among the centrality measures, degree centrality is the com-monly-used one for influence, which is defined as the number of links incident upon a node. Persons who have more ties to others within a network may have more chances to influence others. In an undirected social network, centrality value is the number of connections one person has. With a directed social network, there are two ways to measure centrality: centrality based on out-degree and centrality based on in-degree. Other evolved centrality mea-sures include closeness centrality (considering the distance factor) and betweenness centrality (considering the position factor). In addition, the importance of each reviewer on SNS is also identified by mining the linguistic features (Li et al. 2010). However. These metrics consider the importance of only one node, and the joint influential power of a group of nodes is not accurately reflected by their degree of centrality (Kempe et al. 2003). In contrast, our study discovers the group of users with maximal joint influential power to help companies to conduct online marketing and reputa-tion management.
2.2。
节点的重要性
在社会网络分析中，中心提出测量的重要性的一个节点上，并和中心性较高的节点，然后将大大影响他人（1994年Wasserman和浮士德）。
度中心的中心措施中，通常使用的影响力，它被定义为一个节点后入射的链接数量。
谁拥有更多的关系网络内的其他人可能有更多的机会去影响别人。
在一个无向社会网络，中心值是一个人所具有的连接的数量。
随着向社交网络中，有两种方法来衡量中心性：度和中心度的基础上的中心地位。
其他进化心性MEA确保包括接近中心（考虑距离因素）和中介中心（考虑位置因素）。
此外，SNS上的重要性，每个审稿还确定了挖掘的语言特点（Li等人，2010年）。
然而。
这些度量考虑只在一个节点的重要性，并不能准确地反射度中心性（肯普等人，2003）的联合影响力的一组节点。
相比之下，我们的研究发现最大的联合影响力，帮助企业开展网络营销和声誉管理的用户组。
2.3. Spread of influence
This problem was proposed initially by Pedro Domingos and Matt Richardson in (Domingos and Richardson 2001, Richardson and Domingos 2002), and various studies followed to improve upon this work. The purpose of these studies is to explore the spread of influence in online social networks, and to identify influ-ential users activating other users to buy products. In (Domingos and Richardson 2001), a method based on Markov random field is used to simulate the spread of influence. In (Even-Dar and Shapira 2007, Kempe et al. 2003, Kempe et al. 2005), the spread process of influence is described using the linear threshold model,independent cascade model, and voter model. In (Ben-Zwi et al.2009, Chen et al. 2009, Kimura et al. 2007), a number of efficient approximate algorithms are proposed for this problem. The influ- ence of network structure is explored in (Galstyan et al. 2009);product buying under a competitive social network is analyzed in (Carnes et al. 2007). In addition, other work has focused on diffu-sion and social influence in marketing research. For example, an approach is proposed to identify the specific users who most influ-ence others’ activities for advertising targeting and retention efforts (Trusov et al. 2010), Nair et al. (2010) explore the impact of social interactions and measure the value of influential users in directing targeted sales; a marketing strategy for maximizing revenues is studied in (Hartline et al. 2008).
In contrast to these works, our study focuses on discovering the target group of users with maximal joint influential power, which has high impact on users’ buying, as is evident from the most re-cent findings of marketing research (Katona et al. 2007, 2011).These marketing studies usually assume the condition of the influ-ence spread (such as certain probability), and
based on these assumptions, their methods identify influential users to make as much influence spread as possible. But our method is not based on any assumption, and given the influence strengths between users, can be used with any cases. In addition, these existing stud-ies mainly involve the ‘‘binary value’’ applications, such as buying or not, or adopting or not, while our study encompasses additional‘‘continuous value’’ applications, such as reputation management,brand identification, and public opinion guidance, which require changing users’ impressions of enterprises gradually. For these applications, discovering the group of users with maximal joint influential power is more important because their opinions can do the most to improve an enterprise’s image, strengthen brand recognition, and change public opinion.
2.3。
传播的影响
最初提出这个问题，由佩德罗·多明戈斯和马特·理查德森（2001年多明戈斯和理查德森，理查德森和2002年多明戈斯），并遵循各种研究改进后，这项工作。
这些研究的目的是探索的在线社交网络传播的影响力，并确定有影响的激活其他用户购买产品的用户。
（2001年多明戈斯和理查德森），是用来模拟了一种基于马尔可夫随机场的传播影响力。
在（即使Dar和2007年夏皮拉，肯普等人，2003年，肯普等，2005），传播过程中的影响力是使用线性阈值模型，独立的级联模型，和选民模型来描述。
（al.2009本ZWI等，陈等人2009年，木村等人，2007），这个问题提出一些有效的近似算法。
ENCE网络结构的影响进行了探讨（Galstyan等，2009）;产品购买下一个竞争激烈的社会网络分析（卡恩斯等，2007）。
此外，其他的工作都集中在扩散SION营销研究和社会影响力。
例如，这种方法被提出来识别特定的用户谁最影响ENCE别人的广告活动的目标和保留工作（特鲁索夫等，2010年），Nair等人。
（2010）探讨影响社会交往和测量值有影响力的用户，在指导有针对性的销售收入最大化的营销策略研究（哈特兰等。
2008）。
在这些作品中，我们的研究着重于发现最大的联合影响力，具有高影响用户的购买目标的用户群，是显而易见的，从最重百分之市场调研结果（卡托纳等2007 ，2011）。
这些营销研究通常假设的影响基准蔓延的条件（如一定的概率），并根据这些假设，他们的方法确定有影响力的用户尽可能传播，使尽可能多的影响。
但是，我们的方法还没有任何假设的基础上，给定的用户之间的影响强度，可用于任何情况下。
此外，这些现有的研究主要涉及“二进制值”的应用程序，如购买或不采用或不，而我们的研究包括额外“连续值”的应用，如声誉管理，品牌识别和舆论引导，这需要逐步改变用户对企业的印象。
对于这些应用，发现联合影响力最大的用户群，更重要的是因为他们的意见可以做的最来提高企业的形象，增强品牌知名度，改变舆论。
2.4. Community discovery
Much work has been done on community discovery in socialnetworks (Leskovec et al. 2010). The main purpose has been to partition the nodes of a network into groups, such that the
nodes n one group have better internal connectivity than external con-ectivity. Many proposed principles and algorithms regarding ommunity discovery have been classified into several main cate-gories: spectral algorithms adopt the idea of principal components nalysis to discover communities (Kannan et al. 2004); network ow-based algorithms represent edges with pipes with unit capa-bility, and use max flow-min cut algorithms to identify communi-ties (Flake et al. 2003). Edge-counting ideas (Flake et al. 2000,Radicchi et al. 2004) take a set of nodes with more edges pointing nside the community than do the rest of the network, with some lgorithms using ‘‘betweenness centrality’’ to find community oundaries. Recently, the modularity-based algorithms show otential performances which try to maximize the modularity of iscovered communities (Newman and Girvan 2004). Some studies nfer Web communities from link topology (Girvan and Newman 002); other work analyzes the community structure and semantic community (Gibson et al. 1998, Xu and Chen 2005). In addition, the characters of the social network community are explored, such as small-world property, power-law degree distribution, and network transitivity (Chen et al. 2008, Flake et al. 2002, Zhou et al. 2006). In contrast to these works, the main purpose of our study is to dis-cover a group of users with maximal joint influential power, rather than a group of users that is simply closely connected to eachother.
Our study is different from the existing ones in the following aspects. In contrast to existing studies that explore the importance of a single node in a social network, the objective of this study is to discover the target group of users with maximal joint influential power, which, according to the most recent findings of marketing research (Katona et al. 2007, 2011), is a very important factor influ-encing users’ adoption of products and their holding of positive sen-timental polarities about various companies on SNS. In addition,existing studies usually assume that social relationships between users are symmetric, and an undirected graph is used to model them. However, this is not true for some social relationships, such as the trust relationship. In this study, the directed graph is used to better model the asymmetric influence relationships.
2.4。
社区发现
社区发现在socialnetworks（Leskovec等人，2010年），已经做了许多工作。
的主要目的已分割成组的网络节点，例如，节点n个一组有较好的内部比外部con-ectivity的连通性。
许多建议。
社区发现的原理和算法被分为几个主要类别gories：光谱算法采用的想法的主要成分nalysis发现社区（Kannan等人，2004）;基于网络流算法代表边管单位能力性，并用最大流最小割算法识别通信（片状等，2003）。
边缘计数想法（片状等，2000年，Radicchi等，2004年）一组节点与多个边缘指向内部通讯社区比其余的网络，一些lgorithms使用“中间中心”找到社区oundaries。
近日，的模块化-基础的的的算法显示的otential的表演试试看哪个以最大限度地提高的模块化特性IsCovered该的社区（纽曼至2004年和格文）。
一些的研究NFER链路拓扑（格文和纽曼002）的Web社区其他工作分析群落结构和语义社区（Gibson 等，1998年，徐和2005年陈）。
此外，社交网络社区的字符进行了探索，如小世界，幂律分布，以及网络传递（陈等人，2008年，片状等，2002年，周某等人，2006）。
在这些作品中，我们研究的主要目的是覆盖的用户群最大的联合影响力，而不是一个组的用户，简直是紧密相连的海誓山盟。
我们的研究是在以下几个方面与现有的不同。
在现有的研究，探索在社交网络中的单个节点的重要性相比，本研究的目的是发现目标的用户群最大的联合影响力，其中，根据最新的市场调研结果（卡托纳等，2007年，2011年），是一个非常重要的因素，影响远隔千里的用户通过产品和各种SNS公司持有积极的孙中山timental的极性有关。
此外，现有的研究通常假设用户之间的社会关系是对称的，是用来模拟一个无向图。
然而，这是不正确的一定的社会关系，如信任关系。
在本研究中，有向图来更好的模型的不对称的影响关系。
3. General process for discovering target groups
3.1. Influence network
On SNS, the influence relationships exist only among some sers, and the influence strengths vary between different pairs of sers. In addition, opp osed to the friendship relationship, the influ- nce relationship is asymmetric. Here, the influence network of sers is proposed to describe the influence relationships among sers on SNS represented as a directed graph. A simple influence etwork is shown in Fig. 1The influence network can be represented as G = (V, E, W), and includes the following components:
Node i: represents one user, such as ‘‘vemartin’’ and ‘‘Syntax’’ in ig. 1. The set of all nodes is marked as V = {1, . . . , i, . . . , n}.Arc (i, j): rep resents the influence relationship from user i to j.For example, the arc from ‘‘vemartin’’ to ‘‘Syntax’’ in Fig. 1 shows hat the user ‘‘vemartin’’ has influence on ‘‘Syntax.’’ The set of all arcs is marked as E = {(i, j) e V ⁄ V|i – j}.Weight xij: Each arc is set a weight xij representing the strength of influence; for example, the influence strength from ‘‘vemartin’’ to ‘‘Syntax’’ is 1. The set of all weights is marked as W = {xij|(i, j) e E}. Cost cost(S): There are related costs for targeting a set of users,and the cost of targeting users in S is marked as cost(S).
3。
发现目标群体的一般过程
3.1。
影响网络
SNS的影响关系之间只存在一些SERS，优势各有不同，对不同的SERS的影响。
此外，反对的友谊关系，影响NCE 关系是不对称的。
在这里，影响网络的SERS SERS SNS上表示为一个有向图的影响之间的关系提出了描述。
一个简单的影响ETWORK的图所示。
网络（1）的影响可以表示为G =（V，E，W），并包括下列组件：
节点i：代表一个用户，如“vemartin”和“语法”中的IG。
1。
该组的所有节点被标记为V= {1，。
i。
，n}的圆弧（I，J）：表示从用户i的影响关系到j.For例如，产生的电弧，'' vemartin''''语法''图1显示了帽子的用户vemartin''''语法上的影响力。
所有的弧集为E={（I，J）∈V/ V | I - J}重量XIJ标记：每条弧设置一个XIJ（重量）的强度影响，例如，从“vemartin的影响强度”的“语法”是1。
所有的权重集被标记为W={XIJ|（I，J）E E}。
成本（S）被标记为成本成本（S）：有用于定位的一组用户的相关成本，和S中的目标用户的成本。
In fact, the influence network can represent both directed and undirected influences, for the undirected influence, such as friend-ships, the two influence strengths, wvu and wuv, can be set with the same value.
事实上，影响网络可以代表定向和无向的影响，无向的影响，如朋友，这两个影响的优势，西弗吉尼亚和WUV，可以设置相同的值。
3.2. General process
As we know, there are many SNS, such as Myspace, Twitter, Epi-nions, and so forth. The proposed method can be used for discover-ing the influential groups based on data captured on SNS. Fig. 2 shows the general process of this method:
Data Collection. Some information about user profiles and the influence relationships among users are collected from SNS. This information may include the following:
1) User profile in formation, such as user ID, user role, etc.
2) Social relationships among users, such as a friendship rela-tionship on Facebook and Myspace, a follower relationship on Twitter, and a trust relationship on Epinions. These social relationships usually indica te influence relationships among users as well, because users always more easily believe or accept the opinions of those with whom they have social relationships (Golbeck 2005, Lu et al. 2010, Massa and Avesani 2007).
3) Ratings on users’ reviews. In some S NS, users can rate the opinions of other users. For example, on the Epinions website, users can rate opinions at five levels: ‘‘Not Helpful, Somewhat Helpful, Helpful, Very Helpful, Off Topic’’ to express their assessment of other users’ opinions. These rat ings also indicate the influence relationships among users to some degree: If user A always gives a high rating to user B’s reviews, user A tends to trust user B, and user B will very likely greatly influence user A.
4) Interaction information. On SNS, there are many interactions among users containing supporting or opposing informa- tion; mining these interactions using sentiment analysis and link-based technologies (Pang and Lee 2008) can iden-tify the consistent or inconsistent opinions between users to uncover the use r’s influence relationships.
3.2。
一般过程
正如我们所知道的，有很多的SNS，如MySpace，Twitter的，EPI-nions，等等。
该方法可用于发现SNS上捕获数据的基础上有影响力的群体。
图图2示出了这种方法的一般过程：
数据收集。
SNS的一些信息收集有关用户配置文件和用户之间的影响关系。
此信息可能包括以下内容：
1）用户的个人资料信息，如用户ID，用户的作用，等等。
2）用户之间的社会关系，如友谊关系的关系在Facebook和Myspace，Twitter上的追随者关系，Epinions和信任关系。
这些社会关系通常表明用户的影响之间的关系为好，因为用户总是更容易相信或接受那些与他们有社会的关系（Golbeck2005年，鲁等人，2010年，马萨和2007年Avesani的）的意见。
3）评级在用户的评论。
在一些SNS，用户可以给其他用户的意见。
Epinions网站，例如，用户可以率在五个级别的意见：“没有帮助的，有点帮助，帮助，非常有帮助的，题外话”来表达他们对其他用户的意见。
这些收视率也表明使用者之间的关系，在一定程度上的影响：如果用户A总是给了很高的评价，以用户B的评测，用户à倾向于信任用户B，用户B很有可能会极大地影响用户A
4）交互信息。
SNS上，还有有许多含有信息，支持或反对的用户之间的交互作用;开采这些相互作用利用情绪分析和基于链接的技术（2008年庞李）可以IDEN-TIFY用户之间的一致或不一致的意见，以揭示用户的影响之间的关系。
Influence Network Construction. The influence network can be constructed by analyzing the collected data: The users compose the node set; the influence relationships are built by analyzing so-cial relationships, ratings, and interaction information. The influ-ence strength is set according to the social relationship, the number of positive and negative ratings, and the number of sup-porting and opposing interactions among users. Here, let p be the number of positive ratings and supporting interactions from user v to u, and n be the number of negative ratings and opposing inter-ðpÀnÞactions from v to u; thus, the influence strength is wuv ¼ 1ÀeðpÀnÞ . (For1þe different SNS, site-specific information can be explored to con-struct the influence network; two concrete cases are shown in the experimental evaluation section.)。
