搜档网
当前位置:搜档网 › Monday Report 17

Monday Report 17

Monday Report 17
Monday Report 17

Monday Report第十七期

吕琳媛 曾伟 张千明 王文强

I.学术进展摘要12篇:

NO.1-4曾伟总结,NO.5-8张千明总结,NO. 9-11由王文强总结

II.精读文献4篇:

【1】 Measurement and Analysis of Online Social Networks

………………曾伟(Page6)【2】 Finding Statistically Significant Communities in Networks

………………张千明(Page9)【3】 Using community information to improve the precision of link prediction methods

………………王文强(Page12)【4】 Structural diversity in social contagion

………………王文强(Page13)

I.摘要

1.题目:Beyond Keyword Search: Discovering Relevant Scienti?c Literature

MySpace的朋友推荐系统PYMK

作者:Khalid El‐Arini and Carlos Guestrin

期刊:KDD’11, August 21–24, 2011, San Diego, California, USA.

下载:https://www.sodocs.net/doc/4d13860926.html,/~kbe/bks‐elarini‐guestrin.pdf

平时我们在用google Scholar的时候,输入关键字,系统自动的返回与关键字相关的论文。但是当我们在搜索的时候,我们同时也希望系统能够返回一些不局限于关键字的论文,例如该作者与其他作者合作的论文,后者该领域相关的论文。本文提出了一种新的算法,该算法出了返回与关键字相关的论文以外,还帮助用户找到与论文可能相关的其他论文。实验结果表明,本文的算法明显要好远传统的一些方法。。

2.题目:Measurement and Analysis of Online Social Networks

社会网络分析

作者:Alan Mislove, Massimiliano Marcon, Krishna P. Gummadi, Peter Druschel and Bobby Bhattacharjee

期刊:IMC’07, October 24‐26, 2007, San Diego, California, USA.

下载:https://www.sodocs.net/doc/4d13860926.html,/imc/2007/papers/imc170.pdf

本文研究了四个社会网络Orkut, Youtube, Flickr和LiveJournal,这四个网络都同时包含了用户和用户的朋友关系网络,以及用户加入小组网络,前者一般为friendship,后者一般为membership。本文的研究证实了之前的幂律,小世界以及scale-free等性质。另外本文的研究也证实了这几个网络都具有同配性,即那些度大的节点倾向于链接度大的节点。这篇文章有一些结论值得我们借鉴用于推荐系统。

3.题目:Large‐Scale Matrix Factorization with Distributed Stochastic Gradient Descent

基于矩阵分级的分布式算法

作者:Rainer Gemulla, Peter J. Haas, Erik Nijkamp and Yannis Sismanis

期刊:KDD 2011 August 21‐24, 2011, San Diego, CA.

下载:http://www.mpi‐inf.mpg.de/~rgemulla/publications/gemulla11dsgd.pdf

最近几年,基于矩阵分解算法被广泛应用于推荐系统,该算法除了效率高以外,准确率也要比传统的协同过滤算法好许多。基于矩阵分解算法类似于SVD算法,由于用户和商品的评分数据往往都非常稀疏,基本的SVD算法很难将其分解,一种可行的办法就是利用矩阵的非零项进行分解。一般最小化Frobenius分别得到用户和商品的特征向量。目前有两种办法:一种是alternating-least-squares,另一种是Stochastic Gradient Descent,前者由于在训练的过程中分别固定某个特征向量,因此分布式比较好处理,但是后者由于算法本身的特性,不容易分布式。为了解决这个问题,本文提出了一种新的分布式Stochastic Gradient Descent 矩阵分解算法。实验结果表明,本文的算法效率和准确率都要好于alternating-least-squares 算法。

4. 题目:Predictive Client ‐side Profiles for Personalized Advertising

基于用户信息的个性化广告推广算法

作者:Mikhail Bilenko and Matthew Richardson

期刊:KDD’11, August 21–24, 2011, San Diego, California, USA.

下载:https://www.sodocs.net/doc/4d13860926.html,/apps/pubs/default.aspx?id=157624

本文从客户端的角度,向用户提供个性化的广告推广。本文利用cookie 等技术收集用户的信息,例如用户输入的关键字,浏览过的网页等,从而预测用户以后可能发生的行为,并且向用户提供相应的广告。本文的实验结果表明,向通过提供个性化的广告植入能够帮助提高收入。

5. 题目:Social networks spread rumors in sublogarithmic time

谣言在社交网络中呈sublogarithmic 传播

作者:Benjamin Doerr, Mahmoud Fouz, Tobias Friedrich

期刊:STOC’11 (2011) San Jose, California USA.

下载:https://www.mpi ‐inf.mpg.de/~tfried/paper/2011STOC.pdf

社交网络的流行使得谣言传播的速度剧增。在有关的研究中,经典的PUSH 模型、PULL 模型和PUSH-PULL 模型得出的在PA 模型网络中,谣言传播所需的时间复杂度都是在对数量级,即,本文在这个策略上做了个小小的修改,引入了记忆效应,从而把复杂度降低到了量级,相当于是图的直径。模型修改如下,在谣言传播时,每个点都具有记忆效应(记忆之前传过的路径,容量为M ,取值为0,1,2,3…),而没有选择过的路径则等概率选取。当M 从0变为1,复杂度会急剧下降,但再增大则下降结果不明显了。本文结果可以在一定程度上解释社交网络中“谣言传播速度之快”的问题。

(log )n Θ(log /log log )n Θn 姊妹篇《Why rumors spread so quickly in social networks 》Communications of the ACM 55 (2012) 70。

6. 题目:Controlling edge dynamics in complex networks

复杂网络中连边控制动力学

作者:Tamas Nepusz, Tamas Vicsek

期刊:Nature Physics (2012).

下载:https://www.sodocs.net/doc/4d13860926.html,/pdf/1112.5945.pdf

复杂系统中个体之间的相互作用很自然地上升为复杂系统中的结构特征问题。之前对于此问题的研究已有很多,但是很少有对于网络动力学可控性的研究。基于网络中的连边,本文引入了一个动力学过程,并且证明这个动力学过程的可控性与简单的节点动力学非常不同。实证表明,真实的网络系统比他们的随机化网络更具有可控性;transcriptional regulatory 网络更加容易控制;无标度网络比不相关网络(uncorrelated networks )具有更好的可控性。

7. 题目:On the evolution of user interaction in Facebook

Facebook 中用户的行为演化

作者:Bimal Viswanath, Alan Mislove, Meeyoung Cha, Krishna P . Gummadi

期刊:WOSN’09, 2009, Barcelona Spain.

下载:https://www.sodocs.net/doc/4d13860926.html,/sigcomm/2009/workshops/wosn/papers/p37.pdf

本文抓取了New Orleans区域的Facebook用户的数据,数据的抓取工作分为两个分开的时期。首先在2008年12月29日到2009年1月3日期间仅抓取了用户的关系数据,策略为广度优先搜索,共有90269个用户和3646662条关注关系。然后在2009年1月20日到22日抓取了Fabebook wall的数据,包含838092条信息,包含了每一条信息的时间。这些信息涉及到60290个用户,这些用户通过1545686条边连在一起。本文仅考虑后面的这个子集。统计的结果如下:

1.Facebook墙的日信息量随时间持续增长;

2.基于Facebook墙的信息构建含时间的网络(并不是每条关系的时间都能确立),大

部分用户的行为都很少,少数用户表现活跃:低活跃在加好友一个月后才会去访问

他的wall,通常是生日的祝福来打破僵局;高活跃的用户在加好友之后立刻就会有

交互行为,但频率会随着时间降低;

3.根据wall post的时间信息,生成了网络的9个快照:每到下一个快照都仅有平均

45%的连边留存(如果以每个月为跨度,留存率仅有30%),好在Facebook的增长

速度也很快,所以才有可能使得网络整体的统计特性并没有非常显著的变化。

8.题目:Echoes of power: Language effects and power differences in social interaction

权势的附和:社交行为中言语的影响和权势的差异

作者:Cristian Danescu‐Niculescu‐Mizil, Lillian Lee, Bo Pang, Jon Kleinberg

期刊:2012

下载:https://www.sodocs.net/doc/4d13860926.html,/pdf/1112.3670.pdf

分析在线社区的核心在于理解群体内的社交行为。之前的研究仅局限于结构特性,如谁对谁说话,并且这种行为如何形成更大的网络结构。但是,交互行为自身的形成往往发生在自然语言(或读、或写)形成时,并且也可以合理地认为言语中的信号也能提供发言者的身份、地位等其他信息。因此,本文研究了在群体讨论中,一个人在多大程度上立刻模仿某个人的言语特征,能巧妙地反应出对应参与者的权势大小。基于此现象,本文提出了一个分析框架,用以描述权势差异关系,并且研究在两种不同的情境下(维基百科讨论和美国最高法院中的讨论),谈话行为是如何反映出权势关系的。

9.题目:Discovering value from community activity on focused question answering sites: a case

study of stack overflow

从问题回答网站中的社区活动中发现价值:stack overflow研究案例

作者:A. Anderson, D. Huttenlocher, J. Kleinberg, J. Leskovec

期刊:Proceedings of the 18th ACM international conference on knowledge discovery and data mining 2012 (to be published).

下载:https://www.sodocs.net/doc/4d13860926.html,/people/jure/pubs/sof‐kdd12.pdf

现在有很多问答网站,国外有Quora和Stack Overflow,国内有Baidu知道、知乎等等。

一些十分活跃并且擅长回答某一类问题用户会形成小圈子(community)。他们回答的某些问题也会随着时间的推移变成具有永久价值的问题。有类似疑问的用户可以直接通过搜索引擎搜索到答案。

这篇文章使用Stack Overflow数据研究了问答网站的社区活动,并从中抓取特征来解决两个问题:预测一个问题是不是一个具有长期价值;识别没有被满意回答的问题。结果是都获得了比较高的准确性。从而说明了问答网站中社区活动的意义。

10.题目:Using community information to improve the precision of link prediction methods

使用社区信息提高链路预测方法准确度

作者:S. Soundarajan and J. Hopcroft

期刊:Proceedings of the 21th ACM international conference on World Wide Web 2012, p.605

下载:https://www.sodocs.net/doc/4d13860926.html,/proceedings/companion/p607.pdf

这份工作展示了:使用社区信息辅助结构相似性可以提高链路预测精度。

这份工作从考虑了两方面社区信息:node community和link community。并且试验了两种相似性指标CN和RA。定义CN1为CN加上与待测边的两个端点在同一个社区的共同邻居数量,CNEdge1为CN加上与公共端点是某一个共同邻居且与待测边同在一个link community的共同邻居数量。一次类推RA。发现耦合两种信息的方法叫原来的方法再准确度上均有所提升。

11.题目:Social influence analysis in large‐scale networks

大规模网络社会影响分析

作者:J. Tang, J. Sun, C. Wang and Z. Yang

期刊:Proceedings of 15th ACM international conference on Knowledge Discovery and Data mining 2009, p.807

链接:

https://https://www.sodocs.net/doc/4d13860926.html,/download/attachments/24938081/KDD09_Social_Influence_Analysis.

pdf

这篇文章建立TAP模型(Topical Affinity Propagation),分析了社交网络中的传播过程,并解决了两个问题:如何从一个给定的主题中识别代表性节点,如何识别出某节点的邻居节点对他的影响。这篇文章还使用了Map-reduce模型对TAP模型进行了并行化处理,以便可以处理超大规模的社交网络。

II.精读文献

第一篇(由曾伟总结)

题目:Measurement and Analysis of Online Social Networks

社会网络分析

作者:Alan Mislove, Massimiliano Marcon, Krishna P. Gummadi, Peter Druschel and Bobby Bhattacharjee

期刊:IMC’07, October 24-26, 2007, San Diego, California, USA

链接:https://www.sodocs.net/doc/4d13860926.html,/imc/2007/papers/imc170.pdf

本文研究了四个社会网络Orkut, Youtube, Flickr和LiveJournal,这四个网络都同时包含了用户和用户的朋友关系网络,以及用户加入小组网络,前者一般为friendship,后者一般为membership。本文的研究证实了之前的幂律,小世界以及scale-free等性质。另外本文的研究也证实了这几个网络都具有同配性,即那些度大的节点倾向于链接度大的节点。这篇文章有一些结论值得我们借鉴用于推荐系统。

问题定义:

表1是这几个网络的一些统计特性:

本文的数据是从网站上爬取下来的,所以并没有包含所有的网络数据,但是包含了网络的最大结构(connected component)。

首先是幂律分布,如下图所示:

可以看出,这几个数据节点的度都符合幂律分布,下表是这几个网络对应的幂律指数。

另外下图给出了网络中节点的比例与这些节点的连入边(入度)和连出边(出度)占整个网络边的比例。从图中可以看出,5%的节点连入边占整个网络所有连入边的75%,但是连出边只占了25%。这与微博的情况比较类似,在微博里面,有少许的明星节点,受到大部分人的关注,但是他们关注的人却占了很少部分。

下图给出了top x%节点出度和入度重叠的比例。例如在出度和入度top 1%的节点中,至少有65%是重叠的,换句话说,入度大的节点一般来说,出度也比较大。

下表给出了网络的平均距离和直径,可以看出这些网络偶表现出小世界现象。

下图是当网络中度最大的节点被移除后,网络的链接情况。当去掉网络1%度最大节点,网络中的最大连通结构还存在,但是去掉了10%之后,网络的结构就被完全破坏,剩下许多孤立的节点。

下表给出了网络的聚集系数,可以看出u,这几个网络都表现出高聚集性。

另外,本文也计算了几点度与节点的聚集系数的关系。从下图可以看出,那些度越小的节点,他们的聚集系数越大,而那些度越大的节点,他们的聚集系数反而越小。

第二篇(由张千明总结)

题目:Finding Statistically Significant Communities in Networks

发现网络中具有重要统计性质的社区

作者:Andrea lancichinetti, Filippo Radicchi, Jose J. Ramasco, Santo Fortunato

期刊:PLoS ONE 6 (2011) e18961.

下载:

https://www.sodocs.net/doc/4d13860926.html,/article/fetchObjectAttachment.action?uri=info%3Adoi%2F10.1371%2 Fjournal.pone.0018961&representation=PDF

社区是网络的重要特性之一,然而之前的大部分工作都仅仅考虑社区的单一特性,如聚类、重叠,然而社区还具有“层次性”,而且会出现“假社区”(即,即便是在同度分布的随机化网络中,对应的节点群仍然呈现社区性质)。本文提出的OSLOM(Order Statistics Local Optimization Method)模型,不仅能处理无向网络,还能解决有向、含权网络中的社区划分问题,能同时考虑到社区的可重叠性、层次性以及社区的动力学特性。本方法既可以单独使用,也可以和其他一些社区划分算法结合使用,并且具有较好的效果,更能用于大规模的网络。

该方法的核心思想是“局部聚类统计特性”的最优化问题。本模型的优化过程是以“同度序列的随机化网络(之后简称为随机化的网络)”作为参照对象的,旨在排除“假社区”现象。

首先引入聚类的统计特性(Statistical significance of clusters)。它被定义为“此聚类在随机化的网络中出现的可能性”(假设在随机化的网络中不存在社区结构)。假设从图G开始,节点集合为N和边的集合为E。在G中给定一子图C,一节点i(不在C内),那么剩下的子图可表示为G\[C∪{i}]。如下图所示:

记m C 为子图C 的度,k i 为节点i 的度,如上图m C 加上in 和out 分别表示子图C 的内部的度和外部的连边数,k i 加上in 和out 则分别表示是否指向子图C 。定义子图G \[C ∪{i }]中节点的度数之和为M ,于是M 与子图G \[C ∪{i }]内部度数M *的关系表示如上图所示。

假设C 是随机化网络中的子图,并且子图C 的内部度数固定,如果网络中所有其他边都是随机连接,那么节点i 有个属于C 的邻居的概率为:

*2(|,,)!!()!(/2)in

i k in i out in out in i i C i p k i C G A k k m k M ?=?!

1 其中A 是归一化因子,与无关,用来保证 in i k *:0(|,,)in i in i k M p k i C G ≥=∑

有了这个概率,就可以据此来对所有子图C 以外的节点i 排序。定义下面累积分布

()(|,,i in i k

in i j k r k p j i C G ==)∑ 表示节点i 与子图C 的共同边的数目大于的概率(详细的表达及推导请见[Lancichinetti A, et. al., Phys. Rev. E 81(2010) 046110])。因为节点的度是离散的量,累积分布会有明显的阶梯状,本文为每个节点i 赋予了一个值r ——r i 在[(in i k ),(1)]in in i i r k r k +中随机选取一个数,这个随机数的选取可以通过蒙特卡洛技术实现。

这个变量r 承载了节点与子图C 之间关系的重要信息。拥有最小的r (记为r 1)的节点成为首要考虑的对象,它在随机化的网络中的累积分布则可写作

11()()1(1)C N n r P r r r ?Ω=<=??

其中,n C 是子图C 中的节点数目。

一般来讲,用r q 来表示排在第q 位节点的变量r 的值(升序排列)

,那么累积分布可写作

()()(1)C C N n C N n i i q q i q N n r P r x x x i ???=???Ω=<=??????

∑ 这里采用有序的统计策略是由于“本文假设聚类的方法更倾向于将与社区联系更多的点包含进社区”。不同的表示在社区外的点在多大程度上与随机化网络的期望值相当。为了评估整个社区,在子图C 所有邻居范围内定义了q Ωmin {()}m q q q c r =Ω,其中r q 表示那些邻居节点对应的r 值。c m 的分布依赖于N -n C ,所以累积分布可进一步写为

()(,m C P c x x N n )φ<=?

后文中,将(,)C x N n φ?记为聚类(子图)C 的分值。

接下来,讲述了对于单个聚类(子图)C 的分析。这个部分主要由两步组成:第一步是计算并按条件将子图C 的外邻居节点加入到子图中;第二步是将C 中不满足条件的点移除。这两步合称为“清洗过程”。

第一步:对于C 外的每一个邻居节点i ,计算其对应的r 值,然后找出r 1对应的,并设定阈值P ,如果1()r Ω1((),)C r N n P φΩ?<,就将对应节点加入子图C ;如果1((),)C r N n P φΩ?>,就需要计算第二“最好”的点、第三“最好”的点,以此类推,如果到第q th 个“最好”的点满足了条件,那么就将所有前q 个最好的点加入到子图C 中,此时不再往后计算第q +1个点了;如果没有一个点满足条件,那么就没有新点加入到C ,此时跳过第二步。记更新后的C 为C’。

第二步:对于C’中的每一个点i ,计算其相对于子图C’\{i }的r i ,然后选择最差的点w (即具有最大的r )。然后在子图为C’\{i }的条件下重复第一步,判断节点w 的去留。如果节点w 留下,则结束对此子图的分析;否则,将w 从C’中去除,然后继续寻找C’\{w }中最差的点,重复步骤2。最后得到C *,其中最差的点都被判断为保留。

网络的分析。

上一部分讲述了单个子图C 的分析过程,本部分是将这一过程扩展到整个网络中。首先是随机找一个点i ,于是第一个子图就是C ={i },然后将q 个邻居节点加入到C (q 的取值任意,本文采用3,同幂率分布指数),然后再利用上一步骤中的“单个子图的分析”。

将整个过程在几个不同的初始节点上重复进行(为了减小蒙特卡洛方法带来的误差,也需要多做几次实验),这样一来就会得到一些有重复的社区,当算法探测出的社区比较相似时,算法结束。理想状态下,在不同的初始点下也期望得出完全相同的社区划分,然而实际上几乎不可能。判断两个社区相似的方法可以是|C 1∪C 2|/min(|C 1|,| C 2|)>0.5。

事实上,很多社区都非常相似,那么该如何取舍?

假设C 1和C 2相似,那么是该保留它们还是该保留它们的合集呢?将C 1和C 2在G 3内

做清洗(其中是G 3是C 3的子集)

。如果| C’1∪C’2|>P 2| C 3|,其中P 2设置为0.7,则排除C 3,否则排除C 1和C 2。如果超过两个社区相似,那么方法也是类似。

检测社区下是否有子社区。方法其实很简单,就是给定一子图,在子图里面再按上述的

方法找社区。

所有上述步骤完成之后,输出的结果将是这个网络的覆盖。注意参数P和P2的选取会对结果产生重大影响。若P取小值,社区的规模将会较大;而P2则是控制大社区的区分度。

OSLOM的框架图如页下。

为了加速计算,本方法可以从一个“给定划分”的结果出发进行优化,这也是与其他划分方法一起使用的切入点。使用该方法可以发现:显著的社区、没有归属的点(homeless vertices)、重叠的社区、社区的层次。

此外,该方法还可以通过改进应用到有向网络和含权网络中。

第三篇(由王文强总结)

题目:Using community information to improve the precision of link prediction methods

使用社区信息提高链路预测方法准确度

作者:S. Soundarajan and J. Hopcroft

期刊:Proceedings of the 21th ACM international conference on World Wide Web 2012, p.605. 下载:https://www.sodocs.net/doc/4d13860926.html,/proceedings/companion/p607.pdf

这篇文章使用耦合了社区信息的相似性指标进行链路预测,发现CN和RA的准确度有所提高。

数据集

这篇文章使用了几个常用数据集。Amazon 网络,共同购买同一本书的两个用户之间连接一条边。Facebook学生网络,记载了Rice大学本科生和研究生社交圈子,人类蛋白质网络,酵母菌蛋白质网络,Rovira Virgili大学的电子邮件网络,arXiv两个领域的合作网络,同义词。

实验中,90%的边作为测试集,10%作为测试集。

选用指标

这篇文章使用测试了CN、RA、Jaccard、LHN和Sorensen五种指标,由于CN和RA 的表现最好,文章中只列出了RA和CN的结果。

社区信息

这篇文章利用社区信息的方法是将,在原来相似度指标的基础之上耦合共同社区节点的数量。具体方法如下。

node-community

由于学术界对community还没有严格的数学定义,这里使用了几种生成社区的方法进行比较。对于node community,使用的社区划分方法是louvain method和infomap方法。前者的思想是最大化modularity,后者是基于随机游走对信息流建模划分数据集。经过改造的CN和RA指标记为CN1和RA1,如下所示:

link-community

community是指节点和边反转以后使用距离类方法得到的社区。经过改造后的CN link

和RA记为 CNEdge1和RAEdge1,如下所示:

结果

结果如下所示。

第四篇(由王文强总结)

题目:Structural diversity in social contagion

社会传播中的结构多样性

作者:J. Ugander, L. Backstrom, C. Marlow, and J. Kleinberg

期刊:Proceedings of National Academy of Science in United States of America 109 (20), 7591. 下载:https://www.sodocs.net/doc/4d13860926.html,/content/109/16/5962.short

研究人员往往用疾病传播模型对社交网络上的传播现象进行建模。基本假设是个体被影响的概率随着被感染邻居数目的增加而单调增大。这篇文章在Facebook上对两种传播现象进行了分析,发现个体被感染的概率随着发生联系的邻居(contact neighbor)之间形成的连通子图数目的增大而增大,其绝对数目并不是决定性因素。文章把邻居之间形成的连通子图数量认为是“结构多样性”(structural diversity),因为不同的连通子图可能代表拥有不同社交背景的人群,向个体传递不同的信息。如此,则为该发现提供了一种解释:structural diversity 大的个体认识的人的背景更多样化,多个不同群体的说服力往往比单个群体的更强。

这篇文章研究了两个传播过程,一个是个体收到某个邮件联系人的邀请信而成为Facebook的用户,(user recruitment process)另一个是注册成功的Facebook用户真正开始使用Facebook(use engagement process)。

User recruitment

这篇文章使用相对转化率(relative conversion rate)代表个体被说服注册facebook的可能性大小。首先观测了邻居的数目和邻居连通子图对相对转化率的影响。文章详细列出了2、3、4个邻居的用户的相对转化率的情况。从下面的图中可以看出,固定邻居数量,在连通子图的数目相同时,不同连接情况下的转化率相差不大,但是当连通子图数目增大时,转化率会有相当大的提升。文章将连通子图的数目成为structural diversity。见下图,

另一个可能对转化率产生影响的统计量是连通子图内部的连接密度。文章观测了拥有4、5、6个节点的单个邻居连通子图,发现边的边的数量对转化率没有显著影响。

为了说明邻居的绝对数目并不是主要影响因素,作者观察转化率在连通子图数目一定的情况下随邻居数目增加的变化情况。结果是在连通子图一定的情况下,转化率随邻居数目的增多有下降趋势。

为了进一步说明structural diversity的影响,文章着重讨论了两个邻居的情况。Facebook 上没有好友关系的两个用户在现实生活中也许是好友,这种真实好友的身份可以从Facebook 照片共圈的关系挖掘。作者发现,如果两个邻居没有连边(structural diversity为2),但是曾经共同出现在同一张照片上过,则转化率会降到比有连边的情况(structural diversity为1)更低。如下图所示。

User Engagement

依旧使用转化率对engagement进行衡量。首先作者发现了与recruitment类似的现象,个体被影响的可能性主要受邻居连通子图数目的影响。详见下图。此外作者还观测了再structural diversity相同情况下,发出邀请者的地位(用度衡量)对转化率的影响,发现有如下倾向:结构中成员的度差异越大,转化率越容易受度大邻居的影响。

针对该过程进行的第二项研究是当邻居数目分别为10、20、30、40、50时的个体转化率。但作者发现这几种情况下的邻居往往单独成团,所以作者改考察一下几种结构来定义structural diversity。(1)规模不少于k的连通子图数量。(2)k-core。(3)k-brace。k-brace是所有边的嵌入性(embeddedness)指标都不小于k的子图。边的嵌入性是指边的两个端点的共同邻居数目。详细结果如下图所示。我们发现如果选取足够大的k,diversity对转化率有很好的预测性。

THE END

英语newsreport底稿

校园news播报 主持A 主持B Part:学生街安全问题:情侣A(男)B(女)在逛学生街。刚开始A赞美B多么优雅淑女,B赞美A多么勇猛。突然出现C,D把B拖到一边去,A四周求助,紧张的大叫着:help help robbing robbing。而在几秒钟后,镜头转向B的那边,B勇猛的跟CD搏斗,并成功制服。 主持A 主持B Part:主持人A:问你一个问题,树上有十只鸟,开枪打死一只,还剩几只?主持人B:是无声手枪或是别的无声的枪吗? 主持人A:不是。 主持人B:枪声有多大? 主持人A:会震得耳朵疼。 主持人B:那就是说有80~100分贝? 主持人A:是。 主持人B:在这个城市里打鸟犯不犯法? 主持人A:不犯。 主持人B:您确定那只鸟真的被打死啦? 主持人A:确定。(提问的人已经不耐烦恼了)拜托,你告诉我还剩几只就行了,OK? 主持人B:OK,树上的鸟中有没有聋子? 主持人A:没有。 主持人B:有没有关在笼子里的? 主持人A:没有。 主持人B:边上还有没有其他的树,树上还有没有其他的鸟? 主持人A:没有。 主持人B:有没有残疾的或饿得飞不动的鸟? 主持人A:没有。 主持人B:算不算还在肚子里和卵在鸟窝里的蛋? 主持人A:不算。主持人B:打鸟的人眼有没有花?保证是十只? 主持人A:没有花,就十只。(提问的人已经满头是汗。) 主持人B:有没有傻的不怕死的? 主持人A:都怕死。 主持人B:会不会一枪打死两只? 主持人A:不会。 主持人B:所有的鸟都可以自由活动吗? 主持人A:完全可以。你到底怎么了,我无法跟你合作下去,愤怒离场。。。。。。

主持人B(无辜):如果您的回答没有骗人,(B满怀信心地说,)打死的鸟要是挂在树没掉下来,那么就剩一只,如果掉下来,就一只不剩。 由于主持人愤怒离场,紧急插播广告: 广告内容可以做大的(例如电视上搞推销的一直情绪高昂的推销),也可以做几个小的(可以模仿经典的几个广告) 广告后继续播报 主持C 主持B Part:报道魔方大赛。在比赛现场有两名选手AB。主持人先采访A,然而A技术不佳,越采访越慌乱。在访问了两个问题后终于崩溃,赶主持人走。 主持人尴尬的到B(福龙)那,发现他没动,问:为什么还不开始。B说已经结束,主持人不信,让B重新弄一次。 弄完后可以采访几个问题。 PS:本部分,是为了体现福龙精湛的技术。。。。。。。也算是原创性。 主持C 主持B PART:学生对本校的建议. 不同的建议分为不同的故事场景。 故事1:地点:田家炳楼那边三岔路 A初次来师大想去高区二栋。由于不懂的路,问路人B。路人B说:他不是很清楚,大概是往前走,向左。 A再抓来另一个路人C问,C说:往前走,向右。 最后A怒摔矿泉水瓶说:what the hell,到底怎么走??? 建议:学校该出个本校地图。 故事2:A优雅的拿着一个苹果核。走了很久发现一个垃圾箱,开心的走上前发现上面贴着:坏了,维修中,请到别的地方扔。于是A继续寻找垃圾箱,再过几分钟后发现另一处垃圾箱。A很兴奋的奔跑过去,并且以优雅的姿态想扔进去,这时发现还是坏了。这时A内心os说:没关系,旁边没人会注意。就想直接扔。这时旁边B默默飘过,说了句:请不要乱扔垃圾。A尴尬的走了。 建议:学校应该多弄几个垃圾桶 故事3:A帅哥在篮球场,以一个很帅的姿势灌篮。但因为地板的原因,他滑到了。 建议:学校应该建设一个塑胶篮球场。

BBC News Report 1

News Report (Week One) Neymar injury: Brazil's World Cup dream suffers setback The road that sweeps around the front of this picturesque north-eastern coastal location was populated with fans torn between the joy of quarter-final victory against Colombia and the sadness of seeing their icon and greatest hope left in a local clinic and out of the tournament. Neymar's departure on a stretcher with a fractured vertebra was not exactly a footnote to the win that sets up a semi-final against Germany but few suspected the full consequences of Colombia defender Juan Zuniga's cynical knee in the back at the actual moment of impact. And the scale of Neymar's importance to hosts Brazil and the World Cup was illustrated by the reaction as recognition hit home that the 22-year-old was facing a fight to remain involved in the showpiece that so often bears his name and face here in Brazil. As those inside Fortaleza's EstadioCastelao acclaimed coachLuiz Felipe Scolari and his players at the final whistle after a night of colour, noise and pure theatre, others outside were making their way to where Neymar had been taken for what was ultimately a pessimistic diagnosis. Brazil team doctor Rodrigo Lasmar delivered the news that left the nation fearing their dream of winning the World Cup on home soil had been irreparably damaged by the brutal challenge on Neymar. Scolari, with a typical verbal flourish, announced Neymar had been "hunted" - although plenty could assert with some justification that Brazil had employed similar methods to subdue Colombia forward James Rodriguez. He complained: "Not even a yellow card. Nothing. People think Germany, these or the others, only they are hunted but not Neymar." It was somehow an almost inevitable result of an extraordinarily lenient display by referee Carlos Velasco Carballo, whose refusal to exert his authority on a game that contained more fouls than any other at this World Cup, 54, occasionally threatened a free-for-all and encouraged defenders to take such licence. This figure exceeded Brazil's last game against Chile, when there were 51 fouls. Brazil also conceded more fouls than they have done in any World Cup match since 1966 with 31 while the last time there were more fouls in a single game was when Germany played Argentina in 2006, when 55 were committed. It suggests that for all the commitment to the so-called "La Joga Bonito" ("the Beautiful Game") Scolari's Brazil are prepared to do whatever it takes to win. Whereas the roughed up Rodriguez left the field in tears, with the sort of sympathy from Brazil's players they had pointedly declined to offer him during the game,

两篇news report,双语新闻(附生单词)

Divorce risk for couples sharing chores平分家务的夫妻离婚率高Divorce rates are far higher among “modern” couples who share the housework than in those where the woman does the lion’s share of the chores, a Norwegian study has found. 挪威一项最新调查表明,共同分担家务的“现代”夫妇的离婚率比妻子承担绝大部分家务的夫妇高得多。 In what appears to be a slap in the face for gender equality, the report found the divorce rate among couples who shared housework equally was around 50 per cent higher than among those where the woman did most of the work. 调查报告显示,共同分担家务的夫妇离婚率比妻子承担绝大部分家务的夫妇高出50%。这好似给了性别平等的理念当头一棒。 “What we’ve seen is that sharing equal responsibility for work in the home doesn’t necessarily contribute to contentment,” said Thomas Hansen, co-author of the study entitled “Equality in the Home”. 研究报告的合著者托马斯-汉森说:“从调查可以看出,在家里共同分担家务并不一定会提高婚姻满足感。”这份报告名为《家庭生活中的平等》。 The lack of correlation between equality at home and quality of life was surprising, the researcher said. 汉森表示,夫妻关系的平等和生活质量几乎没有关系,这很令人吃惊。 “One would think that break-ups would occur more often in families with less equality at home, but our statistics show the opposite,” he said. 他说:“人们可能认为一些男女更不平等的家庭离婚率更高,但是结果截然相反。 The figures clearly show that “the more a man does in the home, the higher the divorce rate,” he went on. 而且很明显,“丈夫在家做的家务越多,夫妻离婚率越高。” The reasons, Mr Hansen said, lay only partially with the chores themselves. 汉森说,这与家务事本身仅有部分关系。 The deeper reasons for the higher divorce rate, he suggested, came from the values of “modern” couples rather than the chores they shared. 他表示,双方分担家务的家庭离婚率更高的深层原因来自于“现代”夫妻的价值观,而不是他们分担的家务劳动。 “In these modern couples, women also have a high level of education and a well-paid job, which makes them less dependent on their spouse financially.” Mr Hansen said. 他说:“在现代夫妻中,女性的受教育水平也很高,工作收入也不错,这就降低了她们对于伴侣的经济依赖。” They can mana ge much easier if they divorce,” he said. 他说:“如果她们离婚的话,也能更从容地应对。” Norway has a long tradition of gender equality and childrearing is shared equally between mothers and fathers in 70 per cent of cases. 在挪威,男女平等的观念由来已久。在70%的家庭中,夫妻二人会共同承担抚养子女的义务。 'The world's angriest cat' who has thousands of fans美国“喵上校”被评全球最愤怒的猫 He has won tens of thousands of admirers – but just how many of them would be brave enough to try to stroke him is anyone’s guess.

News report 题目

Unit1 Understanding News Reports

Exercise One Now you will hear two news reports. After each news report, you will hear two or three questions. Listen carefully and choose the best answer from the four possible choices. Report One 1. A. 113. B. 130. C. 120. D. 30. 2. A. To comfort the families of victims. B. To be cautious about this new kind of threat. C. To hang the national flag from their homes and offices by way of homage.

D. To provide some reliable clues about the attack. 3. A. Because they were not satisfied with the government’s pension. B. Because they believed the government should have done better. C. Because they were too sad. D. Because the French media persuaded them not to do so. Report Two 4. A. To find ways of ending child marriage. B. To discuss whether to pass the new bill or not. C. To focus on children’s education and health. D. To convince local traditional leaders and parents. 5. A. 1 6. B. 20. C. 24. D. 18. Exercise Two Listen to the news reports again and complete the missing information to form complete answers to the following questions. Report One 1. Who will attend the solemn ceremony according to the news? The solemn ceremony will be attended by and families of many of those killed. 2. What professions are the victims from? They came from many walks of life, .

高中英语话题8新闻报道(Newsreport)学业水平测试

话题8 新闻报道(News report) Ⅰ.重点单词 1.interview v.采访,会见 2.acquire v.取得;获得;学到 3.inform v.告知;告诉 4.promote v.促进;增进 5.prevent v.阻止,预防 6.threaten v.威胁 7.produce v.生产,产生,出产 8.protect v.保护 9.provide v.提供,供给,供应 10.admit v.承认;许可进入 11.threaten v.威胁 12.waste v. & n.浪费 13.exploration n.考察,探索 14.existence n.生存 15.balance n.平衡 16.appointment n.约会;任命 17.journalist n.记者 18.case n.事实;事例;案件 19.occupation n.工作;占领 20.journalism n.新闻业 Ⅱ.常考短语 1.wealth gap 贫富差异 2.the aging of population 人口老龄化 3.community services 社区服务 4.a low income family 低收入家庭 5.rural left behind children 农村留守儿童6.greenhouse gas 温室气体 7.waste sorting 垃圾分类 8.a potential danger 潜在的危险 9.an alarming increase 惊人的增加 10.social order 社会秩序 11.in harmony with 与……和谐 12.out of control 失去控制 13.cover/go on a story 去采访 14.make an appointment 约会 15.arrange an interview 安排采访 16.have a nose for 对……非常敏感 17.live a low carbon life 过低碳生活 18.raise the awareness of 提高……的意识 19.put a ban on 颁布禁令禁止…… 20.make a joint effort 共同努力

NEWS REPORT1

1.2015-01-12 Up the slope As a leading director of action films who’s adept at(熟练于......) creating spellbinding(adj.引人入胜的v.吸引)visual effects, Tsui Hark surprises us with his latest blockbuster(n.轰动;巨型炸弹;一鸣惊人者), The Taking of Tiger Mountain, which is his first attempt at adapting a revolutionary classic. The 3-D movie is based on Qu Bo’s best-selling novel Tracks in the Snowy Forest. In the 1940s, captain Shao Jianbo (Lin Gengxin) leads a squad(n.班;小队;五人组vt.把......编成班;把......编入班)of around 30 People’s Liberation Army soldiers to wipe out bandits in northeastern China. Seasoned investigator Yang Zirong (Zhang Hanyu) volunteered to go undercover in the bandits’ lair. As a fan of his martial arts films, I am pleased to see the movie still features Tsui’s familiar mythical world of chivalrous men. The protagonist(主角,主演;主要人物,领导者)is both a courageous and resourceful man who places himself at risk in order to protect the local villagers. Every time viewers worry about challenges he faces, he ends up defending himself with ease. However Yang is by no means a flawless(adj.完美的;无瑕疵的;无裂缝的)stereotyped hero. In the beginning, his indiscipline and conceit raised the antipathy of Captain Shao. How the two become tacit partners fighting alongside one another is one of the movie’s highlights. It’s hard to deny that the 3-D in many films in the market is just a selling point (产品特色;销售特色;服务特色). But obviously Tsui has gone much further in using different techniques to make more vivid visual effects than most Chinese directors. In this movie, the slow motion flying bullets and snowflakes all create an immersive experience for the audience.

news report演讲稿

●Good morning, everyone. Today I would like to talk about the first US presidential debate between Hillary Clinton from the Democratic Party and Donald Trump from the Republican Party at the Hofstra University, New York on September 26th. ●My report will be divided into three parts. The first part is the rule of the debate. And the second one is the procedure. The last part will include the result and some comments from media and American netizens. ●Firstly, the rule of the debate. The 90-minute debate with no commercial breaks is divide into six segments, each 15minutes long. It will explore three topic areas: Achieving prosperity; America’s direction; and Securing America. At the start of each segment, the host will ask the same lead-off question to both candidates, and they will each have up to two minutes to respond. From that point until the end of the segment, they’ll have an open discussion. ●Then I’ll show some pictures to you. Staffs at the television station are preparing the show The candidates’ family, Hillary’s husband, also the former president Bill Clinton shakes hands with Donald’s wife Melania Trump. At this moment, as we welcome the candidates: Democratic nominee for president of the United States, Hillary Clinton, and Republican nominee for president of the United States, Donald J. trump. ●Now I’ll introduce t he procedure briefly and choose some key issues. The first round, Achieving prosperity. For taxes, Hillary Clinton wants to address income inequality through increased taxes on the wealthy. She has also called for higher tax breaks for healthcare and education spending for middle-class families. While Mr Trump's current plan includes reducing the number of tax brackets from seven to three, cutting corporate taxes, eliminating the estate tax and increasing the standard deduction for individual filers. And central to that is jobs. Hillary Clinton wants to create jobs by investing in advanced manufacturing, technology, renewable energy, and small businesses. She says independent experts have estimated her plans will create 10 million new jobs. Donald Trump says he will create 25 million jobs over 10 years, saying too many jobs, are being lost to other countries, and suggests that investing in infrastructure, cutting the trade deficit, lowering taxes and removing regulations will boost job creation. ●Then we move into next segment talking about America’s direction. For law and order, Hillary supports laws prohibiting racial profiling and wants to emphasise rehabilitation over long prison sentences for non-violent drug offenders. Trump says law enforcement agencies are unable to fight crime because of runaway "political correctness" and says they should be allowed to get tough on offenders. Guns are a key issue. Hillary Clinton repeats her support for tighter background checks and a ban on assault weapons. She denies the Second Amendment is in peril if she is in

news report

Unit 4 Making the news 新闻报道 感悟范文: (2013江西)星光中学(Xingguang High School)近期举行了一次登山活动。假定你是学校英语报记者,请写一篇短文,报道此次活动。内容包括: 1.时间与地点:4月10日,大青山(Daqing Mountain); 2.活动的过程; 3.你对于这次活动的评论。 注意: 1.词数100左右; 2.可适当增加细节,以使行文连贯。 范文: In order to motivate the students to take outdoor exercise, our school organized a mountain-climbing on April 10. Hundreds of us took part in it. It was a nice day. At round 9 a.m., we gathered at the foot of Daqing Mountain in high spirits, well-equipped for the tough journey to the destination---the top of the mountain. Despite the hardship, we were singing and laughing all the way, enjoying the fresh air and the beautiful views. When some students fell behind or got into trouble, others would come to their assistance. At around 11 a.m., we all arrived at the summit of Daqing Mountain. Bathed in the sunshine, we jumped and cheered with great joy. The activity benefited us a lot. Not only did it get us close to nature and give us relaxation from heavy schoolwork, it also promoted the friendship among us. 【写作点评】 (1)本文用精炼的语言呈现了新闻报道的五大要素; (2)本文是用来过去分词、现在分词作状语以及倒装句式等,体现了用语的多样性。 (3)文章恰当地运用了in order to, in high spirits, well-equipped, despite, come to one’s assistance 等短语,显示了作者的语言功底。 技法讨论:新闻报道是典型的应用文体裁。 一.基本结构: 新闻报道一般分为四个部分:标题、导语、主体、结语。 标题是新闻报道的题目,是对最有新闻价值的内容的浓缩、概括、提炼和再创作。 导语是新闻报道开头的第一段或第一句话,它扼要地揭示出新闻报道的核心内容。 主体是新闻报道的主要部分,它用充分的事实表现主题, 是对导语内容的进一步展开和阐释. 结语一般是最后一句话或者一段话,通常对全文内容作概括性的总结或对新闻事件的发展趋势作出预测。有时, 作者会根据报道的事实在结语中提出令人深思的问题。 二. 注意事项: 1. 搜集素材, 选好报道主题 新闻的主题必须能引起大家的关注和兴趣, 还要具有一定的社会价值,以影响、指导大众的思想、情感、行为为目的。 2. 抓住写作要素 一篇新闻报道要包括五个“W”和一个“H”,即“when, where, who, what, why”和“how”, 即“新闻六要素”。 3. 拟好新闻的标题和导语 标题和导语是新闻的重要组成部分。标题是新闻的眼睛,标题拟得好,可以吸引读者。导语是一篇新闻的第一段或第一句话,它必须具有简明、生动的特点,以帮助读者抓住新闻

英语NEWS REPORT

More elderly Chinese choose to live in nursing homes BEIJING - Qian Liqun, a former professor at the Department of Chinese Language and Literature of Peking University, chose to spend his remaining years with his wife in a nursing home after retirement. As his colleague in Peking University Wen Rumin wrote in his personal Weibo account, Qian would continue his literature research and writing. Though some people questioned the intelligence of the decision by Qian, a famous intellectual, an increasing number of China's senior citizens now choose to live in nursing homes rather than spend their remaining years with their offspring at home. Fortunately for them, modern telecommunications technologies, particularly social networking, has made them feel closer to their children. Gao Lanzhi, 75, lives her life in full without the feeling of loneliness and fear, although her daughters live abroad. One year ago, she moved to the Happiness Senior Citizens Care Center, a private nursing home in west Beijing. Her single room was filled with books and daily necessities. Chinese calligraphy, singing, photography, baking... Gao has many things to do. As she said, "This is my home." She received a knee surgery last year. But after that, she did not hire a nurse to look after her at home. "I cannot bear facing just one

英语版的科技类的news report

Covert Cigarette Butts into Batteries If you see a person carelessly throw a cigarette to the ground, what are you likely to think? That depends. If you hate people throwing things on the ground,you might think, "That person should not litter." If you are a smoker, you might think, "I would sure like a cigarette right now." Most likely, though, you are not thinking, "You know,that unused cigarette butt could be used to make something, something amazing! That is, unless you are a creative scientist. If you are, you may see potential -- a possibility --where others simply see trash. That is just what has happened in South Korea. The spirit of invention can hit at the strangest times and in the most unusual places. In SouthKorea, it happened near a trash can. Kim Gil-Pyo is with the Seoul National University. Mr. Kim says he saw peoplethrowing away cigarette butts. And that got him thinking. He began wondering if something useful could bemade from them. The result? Mr. Kim and other researchers found a way to turn, or convert, cigarette butts into the materials required for high-performance batteries. We often say, "One person's trash is another person's treasure." And in this example, that is unquestionably true. That is, unless you are a creative scientist. If you are, you may see potential -- a possibility --where others simply see trash. That is just what has happened in South Korea. The spirit of invention can hit at the strangest times and in the most unusual places. In SouthKorea, it happened near a trash can. Kim Gil-Pyo is with the Seoul National University. Mr. Kim says he saw people throwing away cigarette butts. And that got him thinking. He began wondering if something useful could be made from them. The result? Mr. Kim and other researchers found a way to turn, or convert, cigarette butts into the materials required for high-performance batteries. We often say, "One person's trash is another person's treasure." And in this example, that is unquestionably true. Super-capacitors are good at storing energy. They have high-power mass, or den sity. Theyrequire only a short time for re-charging. And they have a long lifecycle. Su per-capacitors areused in products such as laptop computers and cell phones. They are also used in industrial energy converters, like wind turbine machines. Combining carbon and super-capacitors seems like a perfect marriage. Kim Gil-Pyo tells the Reuters News Agency that cigarette butts could affect the e conomy is ahuge way. They could prove to be a low-cost source of carbon material. T hey are so cheap thatsmokers throw them to the ground. And you don't get much chea per than that.

相关主题