CAJ | 학술논문

当一个企业或商家需要投放广告时，往往会先通过历史数据、个人兴趣等挖掘出可能购买自己产品的用户，即目标市场（TargetMarket），然后将广告信息通过电视、报纸等公共媒体的形式传递给这些目标用户，希望有更多的目标用户接受信息。然而调查显示，相比于传统大众媒体，人们更倾向于从自己认识的人那里去获取信息，因此文中考虑利用社会影响力的方式去传播广告：在社会网络中说服有限数目的初始用户，并让他们向熟识的人传播信息，期望信息可以通过级联传播覆盖尽可能多的目标用户。由于以往的信息覆盖最大化的工作集中于对全局网络的考虑，因此会忽略目标节点和全局网络之间的联系。通过数据观察可以发现，目标用户往往会由于同质性等原因而聚集在一起，因此文中提出基于聚类的KCC算法，算法通过对用户进行聚类分析，找出每个聚类的代表性用户，使得这些代表性节点可以影响尽可能多的目标用户，同时避免他们之间对信息覆盖的重叠。在不同的真实的数据集的实验显示KCC可以在大多数情况下取得优于其它常用算法的性能，尤其当种子节点数增多时，KCC可以更多地避免节点之间信息覆盖的重叠，从而取得更好的效果；同时，KCC只需要很短的运行时间，具有良好的可扩展性。
당일개기업혹상가수요투방엄고시，왕왕회선통과역사수거、개인흥취등알굴출가능구매자기산품적용호，즉목표시장（TargetMarket），연후장엄고신식통과전시、보지등공공매체적형식전체급저사목표용호，희망유경다적목표용호접수신식。연이조사현시，상비우전통대음매체，인문경경향우종자기인식적인나리거획취신식，인차문중고필이용사회영향력적방식거전파엄고：재사회망락중설복유한수목적초시용호，병양타문향숙식적인전파신식，기망신식가이통과급련전파복개진가능다적목표용호。유우이왕적신식복개최대화적공작집중우대전국망락적고필，인차회홀략목표절점화전국망락지간적련계。통과수거관찰가이발현，목표용호왕왕회유우동질성등원인이취집재일기，인차문중제출기우취류적KCC산법，산법통과대용호진행취류분석，조출매개취류적대표성용호，사득저사대표성절점가이영향진가능다적목표용호，동시피면타문지간대신식복개적중첩。재불동적진실적수거집적실험현시KCC가이재대다수정황하취득우우기타상용산법적성능，우기당충자절점수증다시，KCC가이경다지피면절점지간신식복개적중첩，종이취득경호적효과；동시，KCC지수요흔단적운행시간，구유량호적가확전성。
When an enterprise or company delivers product advertisement into the market,itmight first analyze the purchase history and personal interests of users,and mine valuablecustomers who will potentially buy their products,which is denoted as target marketing.Comparedwith traditional mass media such as newspapers or TV,“word-of-mouth”diffusion in socialnetworks is considered to be more trustful for people.So we consider delivering advertisementwith social influence:by selecting a small set of people as seed nodes to spread information viasocial links,the information of advertisement is expected to cover as many target users as possible.Previous algorithms in influence maximization neglect the relationship between the globalnetwork and local target market,so they are not applicable in this problem.In this paper,wepropose KCC algorithm to select the seed nodes.KCC clusters the nodes and finds representativesin each cluster by defining the similarities and cost of the clusters.It also reduces the overlappingof the coverage of each representative.The algorithm is conducted on different real data sets andcompared with several other well-known algorithms.The results show that KCC performs better in most cases,especially when the seed set size grows larger.Moreover,it only needs a lowrunning time so is scalable for large networks.