1: \documentclass[twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
2:
3:
4: \usepackage{graphicx}% Include figure files
5: \usepackage{dcolumn}% Align table columns on decimal point
6: \usepackage{bm}% bold math
7:
8: \begin{document}
9:
10: \title{A Model of Weighted Network: the Student Relationships in a Class}
11: \author{Bo Hu}
12: \author{Xin-Yu Jiang}
13: \author{Jun-Feng Ding}
14: \author{Yan-Bo Xie}
15: \author{Bing-Hong Wang}
16: \email{bhwang@ustc.edu.cn, Fax:+86-551-3603574.}
17: \affiliation{%
18: Nonlinear Science Center and Department of Modern Physics,
19: University of Science and Technology of China, Hefei, 230026, PR
20: China \\
21: }%
22:
23: \date{\today}
24:
25: \begin{abstract}
26: A simple model is proposed to simulate the evolution of
27: interpersonal relationships in a class. The small social network
28: is simply assumed as an undirected and weighted graph, in which
29: students are represented by vertices, and the extent of favor or
30: disfavor between two of them are denoted by the weight of
31: corresponding edge. Various weight distributions have been found
32: by choosing different initial configurations. Analysis and
33: experimental results reveal that the effect of first impressions
34: has a crucial influence on the final weight distribution. The
35: system also exhibits a phase transition in the final hostility
36: (negative weights) proportion depending on the initial amity
37: (positive weights) proportion.
38:
39: \end{abstract}
40:
41: \pacs{02.50.Le, 05.65.+b, 87.23.Ge, 87.23.Kg}
42:
43: \maketitle
44:
45: \section{Introduction}
46: Recently, physicists have displayed much interests in social
47: phenomena that exhibit complex behaviors and nonlinear dynamics. A
48: social network is a set of people with some pattern of contacts or
49: interactions between them \cite{ref1,ref2}. The patterns of
50: friendships between individuals \cite{ref3,ref4}, business
51: relationships between corporations \cite{ref5,ref6}, and
52: intermarriages between families \cite{ref7} are all examples of
53: networks that have been studied in the past. Hidden behind such
54: complex phenomena, however, are many factors hard to control,
55: including human nature, social environment, social distance and
56: opportunity. Fortunately, human beings have accumulated precious
57: experiences of themselves. Sociologists and psychologists have
58: long noticed that the first impression between two individuals is
59: often the seed of their future relationship, not only seen in
60: romantic stories. Good seed may promise good harvest, while ill
61: seed may be supposed to portend illness. This effect seems to be
62: more distinct in campus life, where the influence of social
63: distances is not so remarkable. Pupils of a class serve as a
64: typical example that exhibits relatively simple friendships. See
65: the studies of friendship networks of school children by Rapoport
66: \cite{ref4}. According to empirical observation, students are more
67: likely to get along with their friends in daily activities, such
68: as dinner, discussion, entertainment, etc. Often, their ties are
69: strengthened through frequent contacts. Assumably, people with
70: common friends or common ``enemies" are prone to unite; likewise,
71: the ``enemy" of Jack's friends or the friend of Jack's ``enemies"
72: may be very difficult to associate with Jack. Similar human
73: relations and social environments are an effective catalyzer for
74: friendships. As a result of restricted social scope and in the
75: light of psychology, the encounters between ``enemies" in a class
76: may also become quite frequent. You cannot avoid your foes in such
77: a small world. It might be an unfriendly eyesight, a provocative
78: action or an unexpected quarrel, as commonplace in daily life. In
79: the deepest of one's heart are always those he hates or he loves,
80: while people without intensified contact are easy to fade from
81: memory.
82:
83: In this paper, a simple model is proposed to study the mechanism
84: of interpersonal relationships within a class. The small social
85: world is assumed as an undirected and weighted graph, where
86: students are represented by vertices, and the depth of favor or
87: disfavor between two are denoted by the weight of corresponding
88: edge. It must be stressed that this model is not restricted in
89: pupil relations, but can be applied to other cases such as
90: interpersonal relations in a club or a team. This paper is
91: organized as follows. In section 2, the model is described on the
92: basis of some simplified assumptions; to understand the mechanism
93: better, we analyze the case for $N$=3. Next, experimental results
94: of the weight distribution are presented and the first impression
95: effect is discussed. In section 4, we conclude with some outlook
96: and possible applications.
97:
98: \section{The Model}
99: The model system consists of $N$ individuals(students of a class).
100: Since the size of a class is not too large, it is suitable to
101: assume that each student has chances to contact all his
102: classmates. For clarity, we introduce a generalization of the
103: $N\times N$ adjacency matrix to describe the interpersonal
104: relationships of the small social network. The matrix elements
105: $\omega_{ij}$ represent the weight of edge $e_{ij}$, where $i$,
106: $j$=1, 2, \ldots, $N$. Postulate that the value of $\omega_{ij}$
107: is discrete and can be negative for the case of disfavor
108: relationships. If most elements of the matrix are positive, the
109: system can be called harmonious; otherwise, it contains
110: considerable hostility. As an original model, we add an assumption
111: that each contact can only alter weight by $\pm1$ at most. In
112: other words, love or hatred is not formed in a day (or individuals
113: will not fall in ``love" at first sight). This condition makes the
114: contacts moderate and can be interpreted by the fact that true
115: friends (or enemies) are selected by time.
116:
117: Now, take $N$=5 for instance, the adjacency matrix is as below:
118:
119: $$
120: \left(%
121: \begin{array}{ccccc}
122: 0 & \omega_{12} & \omega_{13} & \omega_{14} & \omega_{15} \\
123: \omega_{21} & 0 & \omega_{23} & \omega_{24} & \omega_{25} \\
124: \omega_{31} & \omega_{32} & 0 & \omega_{34} & \omega_{35} \\
125: \omega_{41} & \omega_{42} & \omega_{43} & 0 & \omega_{45} \\
126: \omega_{51} & \omega_{52} & \omega_{53} & \omega_{54} & 0 \\
127: \end{array}%
128: \right)
129: $$For undirected and weighted graphs, $\omega_{ii}$=0 and
130: $\omega_{ij}$=$\omega_{ji}$. In the following, we will only
131: discuss the case of symmetric weights. Since the $i$th row of the
132: matrix records the information of interpersonal relationships of
133: student $i$, we will use it to judge the interpersonal similarity.
134: This point is a basic assumption of our model and will be reviewed
135: later. At the beginning of evolution, it is reasonable that some
136: initial weights have non-zero values, due to first impressions
137: among individuals. For convenience, we assign value 1 with
138: probability $p$, and $-1$ with probability $1-p$ to the elements
139: of matrix, i.e. the seed of the model is given. Here, $p$ is
140: called \emph{the initial amity possibility}. The symmetry
141: requirement of the matrix must be satisfied. The initial
142: configuration and the evolution of the system are moderate in
143: degree. The definition of the model is based on the weights'
144: dynamics:
145:
146: (i)First, suppose student $i$ has been randomly selected from the
147: class. Then, he takes the initial to contact student $j$ with a
148: certain probability. A natural idea is to set this possibility as:
149:
150: \begin{equation}
151: P_{i\rightarrow
152: j}=\frac{|\omega_{ij}|}{\sum_{j=1}^N|\omega_{ij}|},
153: \end{equation}recalling the point that ``in the deepest of one's heart are
154: always those he hates or he loves, while people without
155: intensified contact are easy to fade from memory"(see
156: Introduction). However, this method may lead to the absurd case
157: that individual $j$ with $\omega_{ij}$=0 would not be chosen by
158: $i$, nor would $i$ by $j$. Therefore, the edges with 0 weight keep
159: invariant, that is, unfamiliar ones are always unfamiliar. To
160: avoid this unrealistic scenario, let $i$ choose $j$ with
161: possibility:
162:
163: \begin{equation}
164: W_{i\rightarrow
165: j}=\frac{|\omega_{ij}|+1}{\sum_{j=1}^N(|\omega_{ij}|+1)},
166: \end{equation}obviously,
167: \begin{equation}
168: W_{i\rightarrow j}=W_{j\rightarrow i}.
169: \end{equation}
170:
171: The adaption is based on the moderate evolution mechanism: the
172: minimum unit of weight is 1, and each contact can only alter
173: weight by $\pm1$ at most. A non-zero value of $W_{i\rightarrow j}$
174: is needed but we expect a least deflection to $P_{i\rightarrow
175: j}$. When $i$ selects $j$ with possibility $W_{i\rightarrow j}$,
176: for small $|\omega_{ij}|$ this perturbation is quite significant
177: and reasonable. In a fresh environment, people will try to get
178: familiar with others and we call it unfamiliar-familiar period.
179: The differences of weights are not significant; thus, the contacts
180: between them behave no obvious preferences. Once some weight
181: becomes 0 during the period, it is still possible to be altered in
182: later contacts. Initially, the interpersonal relations of the
183: class are unsteady, and the first impressions rising in the period
184: will play an important role in future weight polarization. When
185: friends and enemies (large $|\omega_{ij}|$) form in the system,
186: situation is quite different and we call it friend-enemy period.
187: The contacts between friends and encounters between enemies now
188: become more frequent, and the interpersonal relationships tend to
189: be steady. However, the emergence of new friends and enemies is
190: not forbidden.
191:
192: (ii)Now, $i$ and $j$ have been chosen for interaction, then
193: $\omega_{ij}$ will be altered with a certain possibility:
194:
195: \begin{equation}
196: \omega_{ij}\longrightarrow\omega_{ij}\pm1.
197: \end{equation}
198: The crux of the problem now is how to determine the possibility.
199: Recall that similar human relations and social environments are
200: more likely to promote friendships. We could define $\gamma_{ij}$
201: as below to describe the interpersonal relation similarity:
202:
203: \begin{equation}
204: \gamma_{ij}=C^{-1}\sum_{\alpha}\omega_{i\alpha}\cdot\omega_{\alpha
205: j},
206: \end{equation}where
207: \begin{equation}
208: C=\sqrt{\sum_{\alpha}\omega_{i\alpha}^{2}}\cdot\sqrt{\sum_{\beta}\omega_{j\beta}^{2}}.
209: \end{equation}It is manifest that $\gamma_{ij}$=$\gamma_{ji}$ and
210: $-1\leq\gamma_{ij}\leq$1. One can see that the definition of
211: $\gamma_{ij}$ is equivalent to the inner product of two normalized
212: vectors. So $\gamma_{ij}$ could be regarded as a signed
213: possibility. Then in detail, the rules are: when
214: $\gamma_{ij}\geq$0
215: \begin{eqnarray}
216: \omega_{ij}\rightarrow\omega_{ij}+1,
217: \omega_{ji}\rightarrow\omega_{ji}+1
218: \end{eqnarray}
219: with possibility $\gamma_{ij}$, and nothing is altered with
220: possibility $1-\gamma_{ij}$;
221:
222: when $\gamma_{ij}<$0
223: \begin{eqnarray}
224: \omega_{ij}\rightarrow\omega_{ij}-1,
225: \omega_{ji}\rightarrow\omega_{ji}-1
226: \end{eqnarray}
227: with possibility $|\gamma_{ij}|$, and nothing is altered with
228: possibility $1-|\gamma_{ij}|$.
229:
230: The mechanism (ii) have simple physical and realistic
231: interpretations. Take Jack and Mike for instance. If they have
232: common friends or common enemies, they are more likely to
233: strengthen their friendship ($\gamma>0$). Suppose, however, they
234: are good fellows at first and Jack's pals are all Mike's foes. If
235: Jack goes on associating with Mike, he may be excluded by some of
236: his friends and have to confront his foes under certain
237: circumstances. Thus, their social relations have a potential to
238: separate them ($\gamma<0$), just like an electron-positron system
239: under external electro-magnetic field. In this case, we can
240: equally say that Jack and Mike have distinct social tastes and in
241: the long run, their friendship is on test.
242:
243: After the weights have been updated, the process is iterated by
244: randomly selecting a new individual for the next contact, i.e.
245: going back to step (i) until the class disbands.
246:
247: To better understand the micro dynamics, it is beneficial to
248: analyze the case for $N$=3. Suppose Jack(A), Mike(B) and John(C)
249: interact with each other according to above mechanism. The
250: possible states of this triangle-relation evolution are shown in
251: Fig. 1. Triangle (b) represents the friend-friend-enemy relation,
252: that is, two edges of the triangle have positive(+) weights and
253: another is negative($-$). However, the common friend of
254: antagonistic two will play a conciliatory role in the evolution,
255: and thus, the negative weight will be neutralized at some point,
256: given sufficient interacting time. Review that the ``enemy" of
257: Jack's friends or the friend of Jack's ``enemies" may be hard to
258: associate with Jack (see Introduction). On the basis of similar
259: analysis, triangle (c) and (d) are expectantly the most possible
260: stationary states in the evolution. The asymmetry of the micro
261: mechanism will lead to the asymmetric weight distribution at the
262: macro level. This point will prove itself in the next section.
263:
264: \begin{figure}
265: \scalebox{0.55}[0.55]{\includegraphics{Graph0}}
266: \caption{\label{fig:epsart} the possible states of triangle
267: relationship. Positive(+) edge means friendly relation and
268: negative($-$) represents hostility.}
269: \end{figure}
270:
271:
272: \section{Experimental Results}
273: We choose different \emph{initial amity possibility} $p$ to
274: perform simulations. In order to obtain the weight distribution,
275: the range of weight $\omega$ is equally divided by $M$. Then, the
276: range [$\omega_{min}$, $\omega_{max}$] becomes [$\omega_{1}$,
277: $\omega_{2}$), [$\omega_{2}$, $\omega_{3}$), \ldots,
278: [$\omega_{M}$, $\omega_{M+1}$], where $\omega_{1}$=$\omega_{min}$,
279: $\omega_{M+1}$=$\omega_{max}$. Define $n_{\omega_{l}}$ as the
280: number of weights in [$\omega_{l}$, $\omega_{l+1})$, $l$=1, 2,
281: $\ldots$, $M$; when $l$=$M$ the interval is $[\omega_{M},
282: \omega_{M+1}]$. Plotted in Fig. 2-7 are typical weight
283: distributions ($n_{\omega}\sim\omega$) which behave a pinnacle for
284: $p$=0 or 0.50, power-law with a heavy tail for $p$=0.59 or 0.60,
285: an exponential decay for $p$=0.70, and a peak structure for
286: $p$=1.00.
287:
288: \begin{figure}
289: \scalebox{0.8}[0.8]{\includegraphics{Graph1}}
290: \caption{\label{s:p}weight distribution for N=100, p=0 after
291: $1.0\times10^{6}$ time steps.}
292: \end{figure}
293:
294: When $p$=0, i.e. the initial non-diagonal elements are all $-1$,
295: the final weight distribution exhibits a symmetric pinnacle near
296: $\omega$=0. The peak value is 2690, and on both sides $n_{\omega}$
297: is quite low (but many non-zero), see Fig. 2. The weight
298: distribution for $p\leq$0.50 exhibits a similar behavior, as
299: experiments can test. For $p$=0.50, the peak value is 1693, see
300: Fig. 3. When the initial hostility proportion $1-p$ is
301: significant, the model mechanics has a tendency to push the peak
302: towards the right. However, no matter how many time steps are run,
303: the peak cannot move further beyond zero. We can conclude from
304: these results that when the initial amity is insufficient, the
305: harmony of the class is out of the question. The most majority are
306: indifferent to others, and true friends and foes can rarely
307: ``survive" under such environments.
308:
309: \begin{figure}
310: \scalebox{0.8}[0.8]{\includegraphics{Graph2}}
311: \caption{\label{fig:epsart} weight distribution for N=100, p=0.50
312: after $1.0\times10^{6}$ time steps.}
313: \end{figure}
314:
315: \begin{figure}
316: \scalebox{0.8}[0.8]{\includegraphics{Graph3}}
317: \caption{\label{fig:epsart} weight distribution for N=100, p=0.60
318: after $0.5\times10^{6}$, $1.0\times10^{6}$ and $2.0\times10^{6}$
319: time steps.}
320: \end{figure}
321: Presented in Fig. 4 and Fig. 5 are positive weight distributions
322: for $p$=0.60 and $p$=0.59. The negative weights are discarded in
323: the log-log plot. Apparently, each distribution obeys power law
324: with a heavy tail; this power-law property is independent from
325: iterated steps. Here, negative weights in the matrix are quite
326: sparse, compared with the positive, see Fig. 8 and related
327: discussion.
328:
329: \begin{figure}
330: \scalebox{0.8}[0.8]{\includegraphics{Graph4} }
331: \caption{\label{fig:epsart} weight distribution for N=200, p=0.59
332: after $9.0\times10^{6}$ time steps.}
333: \end{figure}
334: Exponential distributions are found near $p$=0.70 and independent
335: from time steps, as shown in semi-log plot (see Fig. 6). The
336: negative weights have disappeared under such circumstances. It is
337: clear that for large $\omega$, $n_{\omega}$ increases with the
338: passage of time.
339: \begin{figure}
340: \scalebox{0.8}[0.8]{\includegraphics{Graph5}}
341: \caption{\label{fig:epsart} weight distribution for N=100, p=0.70
342: after $0.5\times10^{6}$, $1.0\times10^{6}$ and $2.0\times10^{6}$
343: time steps.}
344: \end{figure}
345:
346: \begin{figure}
347: \scalebox{0.8}[0.8]{\includegraphics{Graph6}}
348: \caption{\label{fig:epsart} weight distribution for N=100, p=1.00
349: after $0.5\times10^{6}$, $1.0\times10^{6}$ and $2.0\times10^{6}$
350: time steps.}
351: \end{figure}
352: When $p$=1.00, i.e. the initial non-diagonal elements are all 1,
353: we find a peak structure in $n_{\omega}\sim\omega$ diagram(Fig.
354: 7). Different from the above, the maximum of $n_{\omega}$ is
355: reached somewhat far from $\omega$=0. By increasing the iterated
356: times, the peak and the upper limit of $\omega$ are both pushed to
357: the right. It means the harmony of the system is boosted up. One
358: can check that for $p\geq$0.8, the weight distribution displays
359: similar behaviors.
360:
361: Three points must be stressed here. First, we have observed four
362: typical kinds of distributions from $p$=0 to $p=1.00$, and each
363: kind appears in a certain range. However, the distributions in
364: some unmentioned ranges are not so typical and may be influenced
365: by increasing time steps. Second, by comparing the weight
366: distributions of different $p$ from 0 to 1, one can see that this
367: system exhibits a potential to become harmonious. Third, the
368: properties of the above weight distributions also suggest a
369: critical transition. Define \emph{the hostility proportion} $h$ as
370:
371: \begin{equation}
372: \sum_{i,j,\omega_{ij}<0}\omega_{ij}\big/\sum_{i,j}|\omega_{ij}|
373: \end{equation}
374: which can describe the harmony degree of the class from an
375: opposite sight. The dependence of $h$ on $p$ is shown in Fig. 8,
376: and a phase transition is found near $p_{c}$=0.6, where the weight
377: distribution exhibits power-law (Fig. 4 and Fig. 5). Below the
378: critical value, the hostility proportion $h$ is non-trivial, that
379: is, there exists considerable hostility in the interpersonal
380: atmosphere; while above the critical value, the final hostility
381: proportion $h$ is close to 0, i.e. the interpersonal atmosphere of
382: the class is quite harmonious. Conflicts and grievances may melt
383: gradually under such harmonious environment.
384:
385: \begin{figure}
386: \scalebox{0.8}[0.8]{\includegraphics{Graph7}}
387: \caption{\label{fig:epsart} the dependence of hostility proportion
388: $h$ on the initial amity proportion $p$, hostility-amity phase
389: transition for N=100 after $1.0\times10^{6}$ time steps.}
390: \end{figure}
391:
392: \section{Review and Outlook}
393: Of the academic disciplines the social sciences have the longest
394: history of the substantial quantitative study of real-world
395: networks \cite{ref9,ref10}. Of particular note among the early
396: works on the subject are: Jacob Moreno's work in the 1920s and 30s
397: on friendship patterns within small groups \cite{ref3}; the
398: so-called "southern women study" of Davis et al. \cite{ref11},
399: which focused on the social circles of women in an unnamed city in
400: the American south in 1936; the study by Elton Mayo and colleagues
401: of social networks of factory workers in the late 1930s in Chicago
402: \cite{ref12}; the studies of friendship networks of school
403: children by Rapoport and others \cite{ref4,ref14}; and the
404: mathematical models of Anatol Rapoport \cite{ref13}, who was one
405: of the most theorists to stress the significance of the degree
406: distribution in networks of all kinds, not just social networks.
407: In more recent years, studies of business communities
408: \cite{ref5,ref15,ref16} and of patterns of sexual contacts
409: \cite{ref17,ref18,ref19,ref20,ref21} have attracted particular
410: attention. However, traditional social network studies often
411: suffer from problems of inaccuracy, subjectivity, and small sample
412: size. Because of these problems many researchers have turned to
413: other methods for probing social networks. One source of copious
414: and relatively reliable data is collaboration networks
415: \cite{ref22,ref23}; another source of reliable data about personal
416: connections between people is communication records of certain
417: kinds \cite{ref24,ref25}. It is quite possible for researchers to
418: investigate the (weighted) relationships in a certain class or
419: club, for its finite size and simple patterns. Previous researches
420: had stressed the significance of the degree distribution in social
421: networks, while more practical studies on the weighted social
422: networks are required.
423:
424: Recently, Alain Barrat, et al. have proposed a general model for
425: the growth of weighted networks \cite{ref26}, considering the
426: effect of the coupling between topology and weights' dynamics. It
427: appears that there is a need for a modelling approach to complex
428: networks that goes beyond the purely topological point.
429:
430: The simple model here is self-generated and allows various further
431: modifications. Some acute interacting ingredients could be taken
432: into this system. For instance, Jack and Mike might be intimate
433: friends long before, so the initial weight between them must be
434: larger. In the present model, moreover, good friendship will not
435: collapse instantly, nor will old grievance; thus it seems more
436: reasonable to take some acute interaction into account. In the
437: framework of this model, it is possible and interesting to study
438: the adaptive process of a new student joining the class midway
439: \cite{ref8}. Meanwhile, this model could be easily extended to
440: directed graph more close to real world where human relations are
441: often asymmetric. Finally, generalizing it to complex social,
442: economic and political networks is also an interesting and
443: challenging task. The relationships between individuals, economic
444: entities or nations are amazingly similar in many aspects. The
445: basic assumptions and concepts here are expected to have well
446: applications in related fields.
447:
448:
449: \begin{acknowledgements}
450: This work was supported by the State Key Development Programme of
451: Basic Research of China (973 Project), the National Natural
452: Science Foundation of China under Grant No.70271070, the
453: China-Canada University Industry Partnership Program (CCUIPP-NSFC
454: No.70142005), and the Doctoral Fund from the Ministry of Education
455: of China.
456: \end{acknowledgements}
457:
458: \begin{thebibliography}{ref1}
459:
460: \bibitem{ref1} J. Scott, {\it Social Network Analysis: A Handbook}, Sage Publications, London,
461: 2nd ed. (2000).
462:
463: \bibitem{ref2} S. Wasserman and K. Faust, {\it Social Network Analysis}, Cambrige University
464: Press, Cambridge (1994).
465:
466: \bibitem{ref3} J.L. Moreno, {\it Who Shall Survive?}, Beacon House, Beacon, NY (1934).
467:
468: \bibitem{ref4} A. Rapoport and W.J. Horvath, {\it Behavior Science 6}, 279-291 (1961).
469:
470: \bibitem{ref5} P. Mariolis, {\it Social Science Quarterly 56}, 425-439 (1975).
471:
472: \bibitem{ref6} M.S. Mizruchi, {\it The American Corporate Network,
473: 1904-1974}, Sage, Beverly Hills (1982).
474:
475: \bibitem{ref7} J.F. Padgett and C.K. Ansell, 1400-1434, {\it Am. J. Sociol. 98},
476: 1259-1319 (1993).
477:
478: \bibitem{ref9} L.C. Freeman, {\it Connections 19}, 39-42 (1996).
479:
480: \bibitem{ref10} J. Scott, {\it Social Network Analysis: A
481: Handbook}, Sage Publications, London, 2nd ed. (2000).
482:
483: \bibitem{ref11} A. Davis, B.B. Gardner and M.R. Gardner,
484: {\it Deep South}, University of Chicago Press, Chicago (1941).
485:
486: \bibitem{ref12} F.J. Roethlisberger and W.J. Dickson, {\it Management and the
487: Worker}, Harvard University Press, Cambridge, MA (1997).
488:
489: \bibitem{ref14} T.J. Fararo and M. Sunshine, {\it A Study of a Biased Friendship
490: Network}, Syracuse University Press, Syracuse (1964).
491:
492: \bibitem{ref13} A. Rapoport, Contribution to the theory of random
493: and biased nets, {\it Bullentin of Mathematical Biophysics 19},
494: 257-277 (1957).
495:
496: \bibitem{ref15} J. Galaskiewicz, {\it Soical Organization of an Unban Grants
497: Economy}, Academic Press, New York (1985).
498:
499: \bibitem{ref16} J. Galaskiewicz and P.V. Marsden,
500: {\it Social Science Research 7}, 89-107 (1978).
501:
502: \bibitem{ref17} P.S. Bearman, J. Moody and K. Stovel, Chains
503: of affection: The structure of adolescent romantic and sexual
504: networks, Preprint, Department of Sociology, Columbia University
505: (2000).
506:
507: \bibitem{ref18} J.H. Jones and M.S. Handcock, An assessment of
508: preferential attachment as a mechanism for human sexual network
509: formation, Preprint, University of Washington (2003).
510:
511: \bibitem{ref19} A.S. Klovdahl, J.J. Potterat, D.E. Woodhouse, J.B. Muth,
512: S.Q. Muth and W.W. Darrow, {\it Soc. Sci. Med. 38}, 79-88 (1994).
513:
514: \bibitem{ref20} F. Liljeros, C.R. Edling, L.A.N. Amaral,
515: H.E. Stanley and Y. Aberg, {\it Nature 411}, 907-908 (2001).
516:
517: \bibitem{ref21} M. Morris, Sexual networks and HIV, {\it AIDS 97: Year in Review
518: 11}, 209-216 (1997).
519:
520: \bibitem{ref22} L.A. Adamic and B.A. Huberman, {\it Science 287}, 2115
521: (2000).
522:
523: \bibitem{ref23} L.A.N. Amaral, A. Scala, M. Barth\'{e}l\'{e}my
524: and H.E. Stanley, {\it Proc. Natl. Acad. Sci. USA 97}, 11149-11152
525: (2000).
526:
527: \bibitem{ref24} W. Aiello, F. Chuang and L. Lu, A random graph
528: model for massive graphs, in {\it Proceedings of the 32nd Annual
529: ACM Symposium on Theory of Computing}, pp. 171-180. Association of
530: Computing Machinery, New York (2000).
531:
532: \bibitem{ref25} W. Aiello, F. Chuang and L. Lu, Random
533: evolution of massive graphs, in J. Abello, P. M. Pardalos, and M.
534: G. C. Resende (eds.), {\it Handbook of Massive Data Sets}, pp.
535: 97-122 Kluwer, Dordrecht (2002).
536:
537: \bibitem{ref26} A. Barrat, M. Barth\'{e}lemy and A. Vespignani, {Phys. Rev. Lett. 92(22)}, 228701 (2004).
538:
539: \bibitem{ref8} B. Hu, X.Y. Jiang, J.F. Ding, Y.B. Xie and
540: B.H. Wang, e-print, cond-mat/0000000, 2004.
541:
542:
543:
544: \end{thebibliography}
545:
546: \end{document}
547: