1 / 54

연결망 분석

우리를 보는 또 하나의 시각. 연결망 분석. 연세 대학교 사회학과 염유식 http://web.yonsei.ac.kr/yoosik/index.htm. 우주 ?.

vinny
Download Presentation

연결망 분석

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. 우리를 보는 또 하나의 시각 연결망 분석 연세 대학교 사회학과 염유식 http://web.yonsei.ac.kr/yoosik/index.htm

  2. 우주? • The movie shows a journey through the simulated universe. On the way, we visit a rich cluster of galaxies and fly around it. During the two minutes of the movie, we travel a distance for which light would need more than 2.4 billion years.

  3. Snapshot of the Universe

  4. Another Universe?

  5. Same look with a little different size August 15, 2006 New York Times SCIENCE ILLUSTRATED; They Look Alike, but There's a Little Matter of Size By DAVID CONSTANTINE One is only micrometers wide. The other is billions of light-years across. One shows neurons in a mouse brain. The other is a simulated image of the universe. Together they suggest the surprisingly similar patterns found in vastly different natural phenomena. DAVID CONSTANTINE Mark Miller, a doctoral student at Brandeis University, is researching how particular types of neurons in the brain are connected to one another. By staining thin slices of a mouse's brain, he can identify the connections visually. The image above shows three neuron cells on the left (two red and one yellow) and their connections. An international group of astrophysicists used a computer simulation last year to recreate how the universe grew and evolved. The simulation image above is a snapshot of the present universe that features a large cluster of galaxies (bright yellow) surrounded by thousands of stars, galaxies and dark matter (web). (Source by Mark Miller, Brandeis University; Virgo Consortium for Cosmological Supercomputer Simulations; www.visualcomplexity.com)

  6. Proteins? Not DNAs Marslov, Sergei and Kim Sneppen Science May 03, 2002 “If you took a given number of proteins and distributed interactions among them randomly, you would hardly find any particular protein that would have a lot of interactions. Proteins would all ‘talk’ randomly with each other in such a network,” Maslov says. “So, hubs of highly-interacting proteins are not something that you would expect to happen by pure chance.” But the scientists did observe hubs of interacting proteins in the yeast cells. The connections between hub proteins reveal an “emergent property” that acts beyond the level of the functions of the individual proteins and makes them act together to coordinate their functions. Studying these interactions can help identify these coordinated functions, and may also reveal intrinsic features of the interacting proteins.

  7. 목차 • 입문 자료 소개 • 왜 연결망인가? 연결망으로 세상보기 • 연결망 자료: 일반 설문자료와는 어떻게 다른가? • 연결망 방법 Network Description Actors Partition Actors’ Position Statistical Approach

  8. 입문 자료 - 1 연결망 분석 소개: 책 • 김용학 2004. 사회연결망이론(개정판). 박영사.  (문화관광부 추천도서) • 김용학. 2003. 사회연결망분석. 박영사.  (학술원 우수도서) • Wasserman, Stanley and Katherine Faust. 1994. Social Network Analysis: Methods and Applications. Cambridge: Cambridge University Press. • Knoke, David and James H. Kuklinski. 1982. Network Analysis. Beverly Hills, California: SAGE Publications. • Scott, John. 1991. Social Network Analysis: a handbook. Newbury Park, California: SAGE Publications. • Wellman, Barry and Berkowitz S.D. 1988. Social Structures: A Network Approach. Cambridge: Cambridge University Press. • Degenne, Alain and Michel Forse. 1999. Introducing Social Networks. London: SAGE Publications

  9. 입문 자료 - 2 연결망 분석 소개: 매뉴얼 • Borgatti, Everett and Freeman. 2004. UCINET 6 User’s Guide. Harvard, MA: Analytic Technologies. • Burt, Ronald. 1991. STRUCTURE Reference Manual. New York, NY: Center for the Social Sciences Columbia University. • Hanneman, Robart A. 2001. Introduction to Social Network Methods. (included in UCINET 6 package)

  10. 입문 자료 - 3 연결망 분석 소개: Internet • http://www.insna.org/

  11. 입문 자료 - 4 연결망 분석 프로그램 • UCINEThttp://www.analytictech.com/ • STRUCTURE http://web.yonsei.ac.kr/yoosik/index.htm • PAJEK http://vlado.fmf.uni-lj.si/pub/networks/pajek/ • 사이람의 NETMINERhttp://www.cyram.com/ • MATLAB, MATHEMATICA

  12. 왜 연결망인가 - 1 속성 vs. 관계망 • 한 개체란 어떻게 정의되는가? • 1950년대 이후의 사회학 이론과 사회학 연구의 괴리: copper + lead  bronze • 여자라서 임금이 작다? 규모가 큰 조직이 혁신이 어렵다?

  13. 왜 연결망인가 - 2 Usual Suspects: 성공한 개인

  14. 개인 vs. 연결망: 존재론으로부터 관계론 유럽 근대사의 근본원리가 근본에 있어서 존재론임에 반해 동양의 사회 구성원리는 관계론. … 근대 사회의 사회론이란 존재론적 세계인식을 전제한 다음 개별 존재들간의 충돌을 최소화하는 사회 질서를 만드는 것. 이에 비하여 관계론적 구성원리는 개별적 존재가 궁극적인 형식이 아니라고 인식. 세계의 모든 존재는 관계망으로 존재. 배타적 독립성이나 개별적 정체성에 주목하는 것이 아니라 최대한의 관계성을 존재의 본질로 규정하는 것. … 관계망은 고전 강독의 화두. - 신영복의 “강의” 2004 -

  15. 개인 vs. 연결망: 인간 ‘인간’은 인간 관계입니다. 德不孤 必有隣 - 論語의 孔子 仁은 무엇인가? 二人 - 論語의 孔子 人間 – 사람과 사람 사이의 관계 인성이란 개념은 어떤 개체나 존재의 속성으로 환원되는 것이라기 보다는 여러 개인이 공동으로 만들어내는 場의 개념 - 신영복의 “강의” 2004 -

  16. 왜 연결망인가 - 3 연결망 분석 예 • Getting a job: Strength of Weak Ties • Performance: Structural Hole • Diffusion: Cohesion vs. Structural Equivalence • Diffusion: Assortative vs. Dissortative

  17. 연결망 자료 - 1 연결망 자료 A B C • Square Matrix • Non-square Matrix

  18. 연결망 자료 - 2 연결망 자료의 종류 • Global Networks:기본적으로 선호모든 종류의 분석 방법 적용 가능random sample? • Ego-centric Network:많은 경우 불가피분석 방법에 제한 • Bi/ Valued Network • Directed/ Un-directed Network

  19. 연결망 자료 - 3 Sampling • Representative Sampling: i.i.d. representative of what? • Snowball Sampling: hidden, small population • Respondent Driven Sampling (Heckathorn)

  20. 연결망 자료 - 4 연결망 분석: 행렬 계산의 예

  21. 연결망 자료 - 5 두 단계 건너서 연결 X =

  22. Network Description- 1 NETWORK DESCRIPTION Figure: Draw Basic: Tools>Univariate Stats Density: Network>Cohesion>Density Diameter: Network>Cohesion>Distance Reachability: Network>Cohesion>Reachability Connectivity: Network>Cohesion>Point Connectivity Transitivity: Network>Cohesion>Transitivity Basic: mean ties, s.d. of ties, etc. across actorsDensity: (actual # of ties)/ (# of all possible ties) Diameter: longest geodesic distance in a network Reachability: 1 if reachable or 0 Connectivity: the number of nodes that would have to be removed in order for one actor to no longer be able to reach another Transitivity: if A  B and B C, then AC

  23. Network Description- 2 자료: 사회 복지 정책과 관련된 10개의 조직 사이의 정보 흐름 (Knoke)

  24. Network Description- 3 Figure: Draw

  25. Network Description- 4

  26. Network Description- 5 Connectivity POINT CONNECTIVITY -------------------------------------------------------------------------------- Input dataset: E:\Program Files\Ucinet 6\DataFiles\KNOKBUR NOTE: This procedure only operates on the first matrix in a dataset. 1 1 2 3 4 5 6 7 8 9 0 C C E I M W N U W W - - - - - - - - - - 1 5 5 3 4 5 1 6 4 4 3 2 5 8 3 5 8 1 6 5 3 4 3 3 3 4 4 3 1 4 3 3 3 4 5 5 3 5 5 1 5 4 3 4 5 5 8 3 5 8 1 6 5 3 5 6 1 1 1 1 1 1 2 1 2 1 7 5 6 3 5 6 1 6 4 2 3 8 5 5 3 5 5 1 5 5 4 4 9 3 3 3 3 3 1 3 3 3 3 10 4 5 3 4 5 1 4 4 3 5 Output actor-by-actor point connectivity matrix saved as dataset PointConnectivity

  27. Actors Partition - 1 Actors Partition: bottom-up • Clique: Network>Subgroups>Cliques • N-clique: Network>Subgroups>N-Cliques • K-plexes: Network>Subgroups>K-Plex • K-cores: Network>Regions>K-Core • Clique: the maximum number of actors who have all possible ties present among themselves. ‘everybody knows everybody’. Maximal complete sub-graph. • N-clique: they are connected to every other member of the group at a distance greater than one. Usually, the path distance two is used. This corresponds to being "a friend of a friend." : 2-clique. Maximal sub-graph where largest geodesic is no greater than n. diameter can be larger than n, and thus not so cohesive group. • N-clans: first identify n-cliques and exclude those n-cliques that have a diameter larger than n • N-clubs: maximal n-diameter graph • K-plexes: a node is a member of a clique of size n if it has direct ties to n-k members of that clique. It requires that members of a group have ties to (most) other group members -- ties by way of intermediaries (like the n-clique approach) do not quality a node for membership. • K-cores: all of whom are connected to some number (k) of other members of the group. The k-core definition is intuitively appealing for some applications. If an actor has ties to a sufficient number of members of a group, they may feel tied to that group -- even if they don't know many, or even most members.

  28. Actors Partition - 2

  29. Actors Partition - 3 Actors Partition: top-down • Components: Network>Regions>Components • Blocks: Network>Regions>Bi-Component • Factions: Network>Subgroups>Factions • Components: sub-graphs that are connected within, but disconnected between sub-graphs. • Blocks: if a node were removed, would the structure become divided into un-connected parts? If there are such nodes, they are called "cutpoints." The divisions into which cut-points divide a graph are called blocks (or bi-components). Identify vulnerable parts. • Factions: ideally, all sub-groups are cliques and each sub-group is component. ‘factions’ produce the closest fractions to this ideal sub-groups. You have to specify the number of factions for the estimation.

  30. Actors Partition - 4 EDUC(3), WRO(6) 4 Factions

  31. Actors’ Position - 1 Actors’ Position • Degree Centrality: Network>Centrality>Degree • Closeness Centrality: Network>Centrality>Closeness • Betweeness Centrality: Network>Centrality>Betweeness • Power: STRUCTURE • Bonacich Power: Network>Centrality>Power • Structural hole: Network>Ego Networks>Structural Holes • Structural Equivalence: : Network>Roles&Positions>Structural • Role Equivalence: STRUCTURE • Brokerage: Network>Ego Networks>Brokerage • Bridgeness: MATLAB PROGRAM

  32. Actors’ Position - 2 1 4 7 2 5 6 8 3 Three Centralities 3/7= 42.857% Degree: # of ties Closeness: 1/ (geodesic distance) Betweeness: 다른 두 점을 이어주는 최단거리에 얼마나 자주 위치하는가? 점 1의 betweeness: 0 점 2의 betweeness: 11: (1-3), (1-5), …, (1-8), (3-2),(3,4),…, (3-8) 7/11= 63.636% 10/21= 47.619% (21=7C2)

  33. Actors’ Position - 3 Burt’s Power: Prominence - 1

  34. Actors’ Position - 4 Burt’s Power: Prominence - 2

  35. Actors’ Position - 5 Burt’s Power: Prominence - 3

  36. Actors’ Position - 6 Bonacich’s Power - 1 Rij는 연결망. β를 정해주면 power를 측정하는 ci가 결정. β의 절대값이 커지면 커질수록 더 멀리있는 행위자에 의해 power가 결정. 만약에 그것이 0이면 모든 행위자의 power는 direct tie에 의해서만 결정. β의 기호는 다른 행위자에 의해 영향을 받는 방향을 결정. 만약에 이것이 음이면 가까운 곳에 힘없는 행위자가 있을 수록 자신의 영향력은 커짐(bargaining). 이것이 양이면, 자신의 영향력도 작아짐. (기존의 측정과 다른 점). α는 다음을 만족하도록 결정됨 (power index를 network size에 상관없이 normalize하는 효과). 따라서 평균 power는 1에 가까움.

  37. Actors’ Position - 7 Bonacich’s Power - 2

  38. Actors’ Position - 8 Bonacich’s Power - 3 • UCINET은 α를 고려하지 않거나 잘못 고려한 값을 계산. From Bonacich’s Paper 만약에 MATLAB 프로그램이 이용 가능하다면 다음과 같은 간단한 함수를 작성하여 Bonacich Power를 계산할 수 있다. “[B] = Bonacich(data set, 0.3)”. It will create a vector, B that contains Bonacich index with a correct adjusted α. =========================================== function [B] = Bonacich(p,beta) le=length(p); % original index without normalizing factor alpha ori = inv(eye(le) - beta*p)*p*ones(le,1); alpha = sqrt (le/ (sum(ori .* ori))); B = alpha * ori; ==================================================== From UCINET’s output

  39. Actors’ Position - 9 Structural Hole - 1

  40. Actors’ Position - 10 Structural Hole - 2

  41. Actors’ Position - 11 Structural Equivalence • two actors are structurally equivalent to the extent that they have identical relations with every other person • 두 사람은 서로 모르는, 심지어는 서로 싫어하는 사이라도 structural equivalent는 가능

  42. Actors’ Position - 12 Role Equivalence - 1 두 행위자가 만약에 같은 종류의 역할을 같은 빈도수로 한다면, 서로 다른 행위자를 대상으로 그 역할을 수행한다 하더라도, 같은 역학을 하는 것으로 간주해야 한다.

  43. Actors’ Position - 13 Role Equivalence - 2

  44. Actors’ Position - 14 Brokerage - 1

  45. Actors’ Position - 15 Brokerage – 2: 제 16대 보건복지 위원회 공동발의의 연결망

  46. Actors’ Position - 16 Brokerage 3 제16대 국회에서는 오로지 조정자와 대리인만 법안 가결에 효과: 중개 대상이 되는 의원 둘 중 적어도 한 명은 같은 당 소속

  47. Actors’ Position - 17 1 4 7 2 5 6 8 3 Bridgeness – 1 (Youm 2007) 1 – 2 – 5 – 4 : 도정 (o), trail (o), 경로 (o)1 – 2 – 5 – 6 – 5 : 도정 (o), trail (o), 경로 (x)1 – 2 – 1 – 2 – 5 : 도정 (o), trail (x), 경로 (x) 경로(path): 정보를 찾고 있는 합리적인 개인들 (최단 거리를 알고 있고 최단거리를 이용하고 싶어한다). Trail: 한번 지나간 행위자를 다시 거치지는 않는다. 소문. 뒷담화. 도정(walk): 병의 전염, 소문의 확산, 조직간의 정보교환 (겸임 이사), 나라나 도시간의 물자 교환/ 관계, 웹 사이트간의 트래픽

  48. Actors’ Position - 18 Bridgeness - 2 • 경로뿐만 아니라 도정까지 고려하는, 또한 부분적인 경로나 도정이 아닌 전체 도정을 전부 고려하는 새로운 측정 방법 • kf*ij: 점 i가 점 k를 거치지 않고 점 j에 도달할 확률. • 만약에 이 확률이 1이라면, 점 i로부터 점j를 이어주는 점 k의 다리 역할은 0이다. 반면에 이 확률이 0이라면, 점 k의 다리 역할은 1이된다:(1 -kf*ij). • kf*ij을 재기위해 (1) 점i로부터 점j로 가는 모든 도정을 밝혀낸다 (2) 그 모든 도정중에서 점 k가 포함되어있는 확률을 구한다.

  49. Actors’ Position - 19 1 4 7 2 5 6 8 3 Bridgeness - 3 Bridgeness는 또한 각 점들 사이의 상관관계도 계산이 가능하게 한다: 점 1과 점 2의 상관관계가 1이면, 어떤 정보나 병이 점1에 도달하면 점 2도 항상 그 정보나 병을 갖게된다. 이 상관관계는 기존의 연결망 분석에서 계산한 것과 다르다: 기존의 것들은 전체 도정이 아닌 (부분적인)경로만 고려했다.

  50. Actors’ Position - 20 Bridgeness – 4: hidden bridges

More Related