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{The Evolution of Interpersonal Relationships in a Social Group}
11: \author{Bo Hu}
12: \author{Xin-Yu Jiang}
13: \author{Bing-Hong Wang}
14: \author{Jun-Feng Ding}
15: \author{Tao Zhou}
16: \author{Yan-Bo Xie}
17: \email{bhwang@ustc.edu.cn, Fax:+86-551-3603574.}
18: \affiliation{%
19: Nonlinear Science Center and Department of Modern Physics,
20: University of Science and Technology of China, Hefei, 230026, PR
21: China \\
22: }%
23:
24: \date{\today}
25:
26: \begin{abstract}
27: A model has been proposed to simulate the evolution of
28: interpersonal relationships in a social group. The small social community
29: is simply assumed as an undirected and weighted graph, where
30: individuals are denoted by vertices, and the extent of favor or
31: disfavor between them are represented by the corresponding edge
32: weight. One could further define the strength of vertices to
33: describe the individual popularity. The strength evolution exhibits
34: a nonlinear behavior. Meanwhile, various acute perturbations to
35: the system have been investigated. In the framework of
36: our model, it is also interesting to study the adaptive
37: process of a new student joining a class midway.
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: A social network is a set of people with some pattern of contacts
47: or interactions between them\cite{ref1,ref2}. Cases that have been
48: studied include the patterns of friendships between
49: individuals\cite{ref3,ref4}, business relationships between
50: corporations\cite{ref5,ref6}, and intermarriages between
51: families\cite{ref7}. In a social group, all individuals will have
52: their friends and enemies. Rapoport and others have once studied
53: friendship networks of school children\cite{ref4,ref14}, due to
54: its relatively simple pattern and small sample size. Recently,
55: Alain Barrat, et al. have proposed a general model for the growth
56: of weighted networks\cite{ref26}, considering the effect of the
57: coupling between topology and weights' dynamics. It appears that
58: there is a need for a modelling approach to complex networks that
59: goes beyond the purely topological point. We have proposed a simple
60: model to simulate the interpersonal relationships among pupils of a
61: class\cite{ref8}. In accordance with empirical observations,
62: similar human relations between people will promote friendship,
63: while opposite interpersonal relations may lead to hostility. The
64: first impression effect, as shown in \cite{ref8}, has played a
65: significant role in the evolution of the interpersonal relations.
66: As a further research and development of our original work, this
67: paper, with both analytical and experimental methods, gives a
68: detailed analysis to the evolution behaviors. The reactions of the
69: system to various perturbations are also investigated.
70:
71: \section{Review The Model}
72: The model system comprises $N$ students and we introduce a
73: $N\times N$ weighted adjacency matrix to describe its
74: interpersonal relationships. The symmetric matrix elements $\omega_{ij}$
75: represent the weight of edge $e_{ij}$, where $i$, $j$=1, 2,
76: \ldots, $N$. The value of $\omega_{ij}$ is discrete and can be
77: negative for the cases of disfavor relationships; each contact
78: between individuals are supposed to alter weight by $\pm1$ at
79: most. This assumption makes the pairwise interaction moderate;
80: thus, love or hatred in this model is not formed in seldom
81: contacts. In the original model, we assign
82: value 1 with probability $p$, and -1 with probability 1-$p$ to the
83: elements of matrix for initialization, where $p$ is called
84: \emph{the initial amity possibility}. The definition of the model
85: is based on the weights' dynamics:
86:
87: (i)First, suppose student $i$ has been randomly selected from the
88: class. Then, he takes the initiative in contacting student $j$
89: with possibility:
90:
91: \begin{equation}
92: W_{i\rightarrow j}=\frac{|\omega_{ij}|+1}{\sum_{j=1,j\neq
93: i}^N(|\omega_{ij}|+1)}.
94: \end{equation}
95:
96: Obviously,
97: \begin{equation}
98: W_{i\rightarrow j}=W_{j\rightarrow i}
99: \end{equation}
100:
101: The reason to choose this possibility is to avoid the scenario
102: that student $j$ with $\omega_{ij}$=0 cannot be selected by
103: student $i$, (for more information, please see Ref\cite{ref8}).
104:
105: (ii)Now, $i$ and $j$ have been chosen for interaction. Since
106: ``birds of a feather flock together" and similar human relations
107: and social environments are likely to promote friendship, we could
108: define $\gamma_{ij}$ as below to describe the interpersonal
109: relation similarity:
110:
111: \begin{equation}
112: \gamma_{ij}=C^{-1}\sum_{\alpha}\omega_{i\alpha}\cdot\omega_{\alpha
113: j}
114: \end{equation}
115:
116: where
117: \begin{equation}
118: C=\sqrt{\sum_{\alpha}\omega_{i\alpha}^{2}}\cdot\sqrt{\sum_{\beta}\omega_{j\beta}^{2}}
119: \end{equation}
120:
121: Apparently, $\gamma_{ij}$=$\gamma_{ji}$ and
122: $-1\leq\gamma_{ij}\leq$1.
123:
124: Then, the rules of revolution are (for more details, please see
125: our original work\cite{ref8}):
126:
127: when $\gamma_{ij}\geq$0
128: \begin{eqnarray}
129: \omega_{ij}\rightarrow\omega_{ij}+1,
130: \omega_{ji}\rightarrow\omega_{ji}+1
131: \end{eqnarray}
132: with possibility $\gamma_{ij}$, and nothing is altered with
133: possibility $1-\gamma_{ij}$;
134:
135: when $\gamma_{ij}<$0
136: \begin{eqnarray}
137: \omega_{ij}\rightarrow\omega_{ij}-1,
138: \omega_{ji}\rightarrow\omega_{ji}-1
139: \end{eqnarray}
140: with possibility $|\gamma_{ij}|$, and nothing is altered with
141: possibility 1$-|\gamma_{ij}|$.
142:
143: After the weights have been updated (*the symmetry requirement of
144: the matrix must be satisfied in every update), the process is
145: iterated by randomly selecting a new individual for the next
146: contact until the class disbands.
147:
148: The most commonly used topological information about vertices is
149: their degree and is defined as the number of neighbors. A natural
150: generalization in the case of weighted networks is the strength.
151: Define strength $\ s_i$ of node $i$ as below to describe individual
152: popularity:
153:
154: \begin{equation}
155: s_i=\sum_{j=1}^N\omega_{ij}.
156: \end{equation}
157:
158: We could write the evolution equation of strength $\ s_i$ as
159: bellow:
160:
161: \begin{eqnarray}
162: \frac{\partial s_i}{\partial t} & = & \sum_{j=1,j\neq
163: i}^{N}\frac{1}{N}W_{i\rightarrow j}\gamma_{ij}+\sum_{j=1,j\neq
164: i}^{N}\frac{1}{N}W_{j\rightarrow i}\gamma_{ji}
165: \end{eqnarray}
166:
167: The first term on the right hand side of Eq. (8) describes all the
168: possible alterations to $s_i$ at time step $t$, contributed by the
169: cases that $i$ is first randomly chosen and then takes
170: initiative in contacting $j$; likewise, the second term describes
171: all contributions from $j$ choosing $i$. Given that
172: $W_{i\rightarrow j}$=$W_{j\rightarrow i}$ and
173: $\gamma_{ij}$=$\gamma_{ji}$, we could simplify master Eq. (8):
174:
175: \begin{eqnarray}
176: \frac{\partial s_i}{\partial t} & = & \frac{2}{N}\sum_{j=1,j\neq
177: i}^{N}W_{i\rightarrow j}\gamma_{ij}\\
178: & = & \frac{2}{N}\frac{\sum_{j=1,j\neq i}^{N}(|\omega_{ij}|+1)\gamma_{ij}}{\sum_{j=1,j\neq
179: i}^N(|\omega_{ij}|+1)}.
180: \end{eqnarray}
181:
182: \begin{figure}
183: \scalebox{0.8}[0.8]{\includegraphics{Graph1}}
184: \caption{\label{s:p}$\gamma$ distribution for p=0.00, N=100 and
185: M=50 after 1.0$\times10^{6}$ time steps.}
186: \end{figure}
187:
188: \begin{figure}
189: \scalebox{0.8}[0.8]{\includegraphics{Graph2}}
190: \caption{\label{fig:epsart} $\gamma$ distribution for p=1.00, N=100 and M=50 after 1.0$\times10^{6}$ time steps.}
191: \end{figure}
192:
193: Weight $\omega_{ij}$ has a certain distribution in the final state
194: (as shown in Ref.\cite{ref8}), so do $W_{i\rightarrow j}$ and $\gamma_{ij}$.
195: For convenience, we define:
196:
197: \begin{equation}
198: <\gamma>_i=\frac{\sum_{j=1,j\neq i}^{N}(|\omega_{ij}|+1)\gamma_{ij}}{\sum_{j=1,j\neq i}^N(|\omega_{ij}|+1)},
199: \end{equation}
200: obviously, $|<\gamma>_i|<$1.
201:
202: To obtain the distribution of $\gamma$, the range
203: of $\gamma$ is equally divided by $M$. Then, the range
204: [$\gamma_{min}$, $\gamma_{max}$] becomes [$\gamma_{1}$,
205: $\gamma_{2}$), [$\gamma_{2}$, $\gamma_{3}$), \ldots,
206: [$\gamma_{M}$, $\gamma_{M+1}$], where $\gamma_{1}$=$\gamma_{min}$,
207: $\gamma_{M+1}$=$\gamma_{max}$. Define $n_{\gamma_{l}}$ as the
208: number of $\gamma$ between [$\gamma_{l}$, $\gamma_{l+1})$, $l$=1,
209: 2, $\ldots$, $M$; when $l$=$M$ the interval is $[\gamma_{M},
210: \gamma_{M+1}]$. As shown in Fig. 1, the distribution of $\gamma$
211: has a pinnacle at $\gamma$=0. The average value of
212: $\gamma$ is $-$0.00118 as calculated. Since now the weight distribution
213: also behaves a pinnacle near $\omega$=0\cite{ref8}, it indicates that when the initial
214: non-diagonal elements(weights) are all $-$1, the system behaves a tardy evolution.
215: Many contacts will give no alteration to the weights, or equally, the effective
216: contacts between people that change their relations occur infrequently.
217: From Fig. 2, one can see that the $\gamma$ distribution exhibits a convexity and reaches the maximum
218: near $\gamma$=0.67. $\gamma>$0.47 and the average of $\gamma$: $<\gamma>$=0.66817.
219: Since the weight distribution $n_{\omega}\sim\omega$ also displays
220: a convexity at $\omega>$0, the effective contacts between individuals are
221: quite frequent. Above results would be reviewed in the next section.
222:
223:
224: \section{Perturbations and Evolutions}
225: In the original model, we suppose that individuals
226: at first are not familiar with each other.
227: The initial configuration of the system
228: is determined by the initial amity possibility
229: $p$. It can be interpreted as the first impression among them.
230: However, it is completely possible that two individuals have
231: good fellowship or a deep grievance previously.
232: Plotted in Fig. 3 is $\omega_{0}\sim t$ for the process of an old grievance
233: thawing in an initial harmonious group ($p\approx$1.00). $\omega_{0}$, which
234: denotes the old grievance between two from the very beginning,
235: rises sharply from -100 to 0, and increases gently afterwards.
236: This phenomenon can be interpreted from two aspects. On the one hand,
237: the new environment provides them sufficient opportunities to contact other people,
238: so they would not encounter frequently and aggravate their greivance.
239: On the other hand, they both will get on well quickly with new individuals,
240: due to $p\approx$1.00 (see Ref.\cite{ref8}). Thus, after sufficient contacts
241: they will have common friends who will play a conciliatory
242: role in mitigating their old grievance.
243:
244: \begin{figure}
245: \scalebox{0.8}[0.8]{\includegraphics{Graph6}}
246: \caption{\label{fig:epsart} weight evolution for N=100 after 1.0$\times10^{7}$ time steps.
247: Initially, all the weights are set +1, except $\omega_{0}$=-100, where $\omega_{0}$
248: denotes the designated negative weight.}
249: \end{figure}
250:
251: All too often, a group has its leader who has great
252: reverence among his people. He may be an elder chief, a captain
253: of a football team or a chairman of an association. The other members of
254: the group may have some conflicts and grievances. We are interested in the
255: role of the leader playing in the development of the collective.
256: For simplicity, we could assign a large integer to the matrix elements
257: at row $i$ ($\omega_{ij}$=$\omega_{ji}$ and $\omega_{ii}$=0) to
258: denote the leader's relations, and set all the other non-diagonal
259: elements as -1. In Fig. 4, we have taken this integer as 100. Apparently, the leader's strength
260: grows linearly with the passage of time, while the strength of his members increases with
261: a relatively slow velocity. By comparison with the evolution containing no leader, one
262: could conclude that the strengths of the system have been boosted up by the leader. The slope of the
263: leader's popularity evolution is 0.03167. In section 2, we have given that:
264: \begin{equation}
265: \frac{\partial s_i}{\partial t}=2<\gamma>_i/N.
266: \end{equation}
267: For the case containing the leader: if we assume that $\omega_{kj}\ll\omega_{ij}$ for all $j$ and $i\neq k$,
268: then $\gamma_{ij}\approx$1, and further $\partial s_i/\partial t$=2/$N$=0.04, in approximate agreement
269: with the experimental results. For the case absent the leader:
270: the average speed of their popularity growth can be also estimated.
271: If we suppose $<\gamma>_i\approx<\gamma>$=$-$0.0118 (as mentioned in section 2),
272: then Eq. (12) gives the average velocity of popularity growth $-$7.2$\times10^{4}$,
273: which is quite low. But as one can see in Fig. 4, their strengths are dispersed
274: at a certain time (nonlinear property). Note that the master
275: equation (8) can only give an average rate of popularity growth.
276: The leader has changed the possible state of disunity and make his men draw together.
277:
278: \begin{figure}
279: \scalebox{0.8}[0.8]{\includegraphics{Graph4}}
280: \caption{\label{fig:epsart} strength evolution for N=50, M=50 and p=0.00 after 2.0$\times10^{7}$ time steps.
281: The dot line (with slop 0.03167) fits the strength evolution of the leader, and
282: blacksquare($\blacksquare$) represents the strength of
283: others, circle($\circ$) denotes the strength evolution without the leader, under the same conditions. The data
284: are drawn out at intervals of 4.0$\times10^{6}$ time steps.}
285: \end{figure}
286: It is also interesting to study the opposite case. We could put a public
287: anti into a united group to see the strength evolution of the system, see Fig. 5.
288: For initialization, we assign $-$100 to the matrix elements at
289: row $i$ ($\omega_{ij}$=$\omega_{ji}$ and $\omega_{ii}$=0) to represent the anti's relations,
290: and set all the other non-diagonal elements as 1 to describe the initial public.
291: Obviously, the anti's popularity decreases linearly,
292: while the strength of the public rises at a relatively slow speed
293: (approximate linear). By comparing with the case with no anti,
294: one could see that the strength evolution
295: of the public has not been changed significantly by the anti.
296: The anti's incompatibility to the public is aggravated.
297: Recall that $<\gamma>$=0.67 in section 2, and suppose $<\gamma>_j\approx<\gamma>$ (for $j\neq i$).
298: We could estimate the growth rate of the public strength
299: as $2<\gamma>_j/N\approx2<\gamma>/N$=$2\times0.76/50$=0.0304.
300: The slope of the anti's popularity evolution is $-$0.0425.
301: It is reasonable to assume that $|\omega_{kj}|\ll|\omega_{ij}|$ for all $j$ and $i\neq k$,
302: then $\gamma_{ij}\approx-$1, and further $\partial s_i/\partial t$=2/$N$=$-$0.04,
303: in fine agreement with the experimental results.
304:
305:
306: \begin{figure}
307: \scalebox{0.8}[0.8]{\includegraphics{Graph5}}
308: \caption{\label{fig:epsart} strength evolution for N=50, M=50 and p=1.00 after 2.0$\times10^{7}$ time steps.
309: The dot line (with slop -0.0425) fits the strength evolution of the public anti, and
310: blacksquare($\blacksquare$) represents
311: the strength of the public, circle($\circ$) denotes the strength evolution without the anti, under the same
312: conditions. The data are drawn out at intervals of 4.0$\times10^{6}$ time steps.}
313: \end{figure}
314:
315: It must be stressed that the strength growth is actually
316: nonlinear. As Fig. 6 indicates, the trace of the strength
317: evolution behaves linear initially, but nonlinear afterwards.
318: Experiments demonstrate that the strength evolution exhibits
319: obvious nonlinear behaviors after sufficient time, whatever the
320: initial state. When $p$=1.00, it has a widest linear zone, but
321: finally, it splits. We would like to discuss the adaptive process
322: of a student joining a class midway. Plotted in Fig. 8 (log-log
323: scale) is the strength evolution of the system on the background
324: of $p$=1.00. The old students' popularity grows linearly, even
325: after inserting the new one. One could see the nonlinear growth
326: for the new-coming pupil, whose popularity grows approaching the
327: asymptote of the old students' strength evolution. The adaptive
328: process for the new is quite short, thanks to the harmonious
329: group.
330: \begin{figure}
331: \resizebox{0.95\columnwidth}{!}{\includegraphics{Graph3}}
332: \caption{\label{fig:epsart} strength evolution for N=10, p=0.00 after 2.0$\times10^{7}$ time steps.}
333: \end{figure}
334:
335:
336: \begin{figure}
337: \scalebox{0.7}[0.7]{\includegraphics{Graph7}}
338: \caption{\label{fig:epsart} strength evolution of a student joining midway: N=49 and p=1.00. The new student
339: joins the class after 4.0$\times10^{5}$ time steps, and his evolution is denoted by $\diamondsuit$.
340: the strength evolution of old students are denoted by $\blacksquare$. We have extracted 5 from them as
341: representatives.}
342: \end{figure}
343:
344:
345:
346:
347: \section{Outlook and Applications}
348: This paper is focused on discussing various perturbations and corresponding evolutions
349: of our original model, which is proposed to study the student relationships
350: in a class. Both experimental and analytical methods have been used to learn human relations in a
351: local group. The generalization of the model to describe the real social networks
352: is still a challenging problem.
353: (to be continued.)
354:
355: \begin{acknowledgements}
356: This work was supported by the State Key Development Programme of
357: Basic Research of China (973 Project), the National Natural
358: Science Foundation of China under Grant No.70271070, the
359: China-Canada University Industry Partnership Program (CCUIPP-NSFC
360: No.70142005), and the Doctoral Fund from the Ministry of Education
361: of China.
362: \end{acknowledgements}
363:
364: \begin{thebibliography}{ref1}
365: \bibitem{ref1} Scott, J., {\it Social Network Analysis: A Handbook}, Sage Publications, London,
366: 2nd ed. (2000).
367:
368: \bibitem{ref2} Wasserman, S. and Faust, K., {\it Social Network Analysis}, Cambrige University
369: Press, Cambridge (1994).
370:
371: \bibitem{ref3} Moreno, J. L., {\it Who Shall Survive?}, Beacon House, Beacon, NY (1934).
372:
373: \bibitem{ref4} Rapoport, A., and Horvath, W. J., A study of a large
374: socialgram, {\it Behavior Science 6}, 279-291 (1961).
375:
376: \bibitem{ref5} Mariolis, P., Interlocking directorates and
377: control of corporations: The theory of bank control, {\it Social
378: Science Quarterly 56}, 425-439 (1975).
379:
380: \bibitem{ref6} Mizruchi, M. S., {\it The American Corporate Network,
381: 1904-1974}, Sage, Beverly Hills (1982).
382:
383: \bibitem{ref7} Padgett, J. F. and Ansell, C. K., Robust action and
384: the rise of the Medici, 1400-1434, {\it Am. J. Sociol. 98},
385: 1259-1319 (1993).
386:
387: \bibitem{ref8} B. Hu, X.-Y. Jiang, J.-F. Ding, Y.-B. Xie. and
388: B.-H. Wang, Preprint, cond-mat/0408125.
389:
390:
391: \end{thebibliography}
392:
393: \end{document}
394: