1: \documentclass[a4paper, 12pt]{article}
2: \usepackage{epsfig}
3: \usepackage{graphicx}
4: \usepackage{amsmath}
5: %documentstyle[12pt,epsfig]{article}
6: %\oddsidemargin .5cm
7: %\evensidemargin .5cm
8: %\textheight 21truecm
9: %\textwidth 15truecm
10:
11: \author{
12: Juli\'an Candia$^{a,b}$, Marta C. Gonz\'alez$^{a,b}$, Pu Wang$^{a,b}$,\\
13: Timothy Schoenharl$^{c}$, Greg Madey$^{c}$, Albert-L\'aszl{\'o} Barab\'asi$^{a,b,d}$\\{}\\
14: $^a${\small\it Center for Complex Network Research and Department of Physics,}\\
15: {\small\it Northeastern University, Boston, MA 02115, USA}\\
16: $^b${\small\it Department of Physics, University of Notre Dame, Notre Dame, IN 46556, USA}\\
17: $^c${\small\it Department of Computer Science and Engineering,}\\
18: {\small\it University of Notre Dame, Notre Dame, IN 46556, USA}\\
19: $^d${\small\it Collegium Budapest, Szenth\'aroms\'ag u. 2, H-1014 Budapest, Hungary}
20: }
21:
22: \title{Uncovering individual and collective human dynamics from mobile phone records}
23:
24: \begin{document}
25: \maketitle
26:
27: \begin{abstract}
28: Novel aspects of human dynamics and social interactions are investigated by means of
29: mobile phone data. Using extensive phone records resolved in both time and space, we
30: study the mean collective behavior at large scales and focus on the occurrence of anomalous events.
31: We discuss how these spatiotemporal anomalies can be described using standard percolation theory tools.
32: We also investigate patterns of calling activity at the individual level and show
33: that the interevent time of consecutive calls is heavy-tailed. This finding, which has
34: implications for dynamics of spreading phenomena in social networks, agrees with results previously
35: reported on other human activities.
36: \end{abstract}
37:
38: \section{Introduction}
39: Mobile phones are becoming increasingly ubiquitous throughout large
40: portions of the world, especially in highly populated urban areas and
41: particularly in industrialized countries, where mobile phone penetration
42: is almost $100\%$.
43: Mobile phone providers regularly collect extensive data about the call volume, calling
44: patterns, and the location of the cellular phones of their subscribers. In order for
45: a mobile phone to place outgoing calls and to receive incoming calls, it must
46: periodically report its presence to nearby cell towers, thus registering its position
47: in the geographical cell covered by one of the towers.
48: Hence, very detailed information on the spatiotemporal
49: localization of millions of users is contained in the extensive call records of any
50: mobile phone carrier. If misused, these records - as well as similar datasets on
51: buying habits, e-mail usage, and web-browsing, for instance - certainly pose a serious
52: threat to the privacy of the users. However, the use
53: of privacy-safe, anonymized datasets represent a huge scientific opportunity to uncover the structure and
54: dynamics of the social network at different levels, from the small-scale individual's
55: perspective to the large-scale, collective behavior of the masses, with an unprecedented
56: degree of reach and accuracy.
57: Besides the inherent scientific interest of these issues,
58: deeper insight into applications of great practical importance could certainly be gained. For instance,
59: urban planning, public transport design, traffic engineering, disease outbreak control, and
60: disaster management, are some areas that will greatly benefit from a better understanding
61: of the structure and dynamics of social networks \cite{gon07}.
62:
63: The use of mobile phone data as a proxy for social interaction has already proved
64: successful in several recent investigations. Onnela {\it et al.} \cite{onn07a,onn07b} have
65: analyzed the structure of weighted call graphs arising from reciprocal calls
66: that serve as signatures of work-, family-, leisure- or service-based relationships. A coupling between
67: interaction strengths and the network's local structure was observed, with the counterintuitive consequence
68: that social networks turn out to be robust to the removal of the strong ties but fall apart following a
69: phase transition if the weak ties are removed. Szab\'o and Barab\'asi \cite{sza06} have studied social network
70: effects in the spread of innovations, products and new services. They investigated different mobile
71: phone-based services and found the coexistence on the same social network of two distinct usage classes,
72: with either very strong or very weak community-based segregation effects.
73: In the context of urban studies and planning, Ratti {\it et al.} \cite{ratti1,ratti2} have considered
74: the potential use of aggregated data from mobile phones and other hand-held
75: devices. Their ``Mobile Landscapes"
76: project aims at the application of location based services to urban studies in order to gain insight into
77: complex and rapidly changing urban dynamics phenomena.
78: More recently, Palla, Barab\'asi and
79: Vicsek \cite{pal07a, pal07b} used mobile phone data to study the evolution of social groups. They found that
80: large groups persist for longer times if they are capable of dynamically altering their membership, suggesting
81: that an ability to change the group composition results in better adaptability. In contrast, the behavior of small
82: groups displays the opposite tendency, the condition for long-term persistence being that their composition remains stable.
83:
84: In the following sections, we present new results that
85: address novel aspects of human dynamics and social interactions obtained from extensive mobile phone data.
86: In Sect. 2 we show how large-scale collective behavior can be described using aggregated data
87: resolved in both time and space. We stress the importance of investigating large departures from the average and
88: develop the basic framework to quantify anomalous fluctuations by means of standard percolation theory tools.
89: In Sect. 3 we focus on the individual level and study patterns of
90: calling activity. We show that the interevent time of consecutive
91: calls is heavy-tailed, a finding that has implications for the
92: dynamics of spreading on social networks~\cite{pas01,Zoltan,Grenfell:Science,Vespignani,gon_epi1,gon_epi2,
93: can06,can07a,can07b}. Furthermore, by
94: fixing the time of observation between consecutive
95: calls it is possible to use the phone call data
96: to characterize some aspects of human mobility.
97:
98: \section{Fluctuations in aggregated spatiotemporal call activity patterns}
99:
100: The spatial dependence of the call activity at any given time can be conveniently displayed by
101: means of maps divided in Voronoi cells, which delimit the area of influence of each transceiver tower or antenna.
102: The Voronoi tessellation partitions the plane into polygonal regions,
103: associating each region with one transceiver tower. The partition is such that all points within a
104: given Voronoi cell are closer to its corresponding tower than to any other tower in the map.
105:
106: Figure 1 shows activity maps for aggregated data corresponding to a 1-hour interval. The upper panel
107: shows the activity pattern (in log$_{10}$ scale) for a peak hour (Monday noon),
108: while the lower panel shows the same urban neighborhood
109: during an off-peak hour (Sunday at 9 am). The differences between both panels reflect the intrinsic rhythm
110: and pulse of the city: we can expect call patterns during peak hours to be dominated by the hectic activity
111: around business and office areas, whereas other, presumably residential and leisure areas can
112: show increased activity during off-peak times, thus leading to different, spatially distinct activity patterns.
113: Besides different spatial patterns, each particular time of the day, as well as each day of the week,
114: is characterized by a different overall level of activity. This phenomenon is shown by the plot at the center
115: of Figure 1, in which aggregated data for a country is shown as a function of time (data was binned
116: in time intervals of 1 hour). As expected, the overall normalization of the aggregated pattern is lower
117: during weekends than during weekdays, except around weekend midnights and early mornings, when many people go out.
118:
119:
120: \begin{figure}[t!]
121: \includegraphics[width=5.8truein, height=4.2truein]{fig1.eps}
122: \caption{Call activity maps in an urban neighborhood, showing the number of calls per hour
123: managed by each transceiver tower or antenna (dots). The division in terms of Voronoi cells defines
124: the area of reach of each tower. Call traffic patterns depend on time and day of the week, as shown by
125: comparing the map on a Monday at noon (upper panel) with that on a Sunday at 9 am (lower panel). The bars
126: on the right side of each panel correspond to the number of calls per hour and tower in log$_{10}$ scale.}
127: \label{fig1}
128: \end{figure}
129:
130: The minimum spatial resolution is determined by either the typical distance between towers or, in
131: rural regions with sparse tower density, by the reach of the radio-frequency signals exchanged between the
132: mobile handset and the antenna (typically ranging from a few hundred meters to several kilometers).
133: To explore activity differences at larger scales, the data of neighboring cells can be aggregated.
134: At the expense of some loss of spatial resolution, aggregating data into larger spatial bins (taking, e.g.,
135: a regular spatial grid covering the entire country) allows for better statistics and for a more stable
136: activity pattern. That is, the number of calls made from a group of nearby cells at a certain time
137: and day of the week is expected to be fairly constant, except for small statistical fluctuations.
138:
139: Usually, activity patterns are strongly correlated with the daily pulse of populated areas (such
140: as those shown in Fig. 1) and, at a larger scale, to variations in population density between different
141: regions within the country. In contrast, departures from the mean expected activity are in general not
142: trivially correlated with population density and describe instead interesting dynamical features.
143:
144: The measurement of fluctuations around the mean expected activity is of paramount importance, since
145: it allows a quantitative measurement of anomalous behavior and, ultimately, of possible emergency
146: situations. This indeed constitutes the base of proposed real-time monitoring tools such as the
147: {\it Wireless Phone-based Emergency Response} (WIPER) system \cite{mad06}. Anomalous patterns indicative of
148: a crisis (such as the occurrence of natural catastrophes and terrorist attacks) could be detected
149: in real time, plotted on satellite and GIS-based maps of the area, and used in the immediate evaluation of
150: mitigation strategies, such as potential evacuation routes or barricade placement, by means of computer
151: simulations \cite{mad06,sch07}.
152:
153: \begin{figure*}[t]
154: \begin{center}
155: \includegraphics*[width=10.0cm]{fig2.eps}
156: \end{center}
157: \caption{Activity and fluctuations in a regular 2D grid showing a normal event (left panels) and an
158: anomalous one (right panels). The activity is displayed in terms of the number of calls
159: per hour inside each square bin in log$_{10}$ scale (upper panels). High-activity bins above the
160: fluctuation threshold $A_{thr}=0.25$ are shown in black, while bins with normal activity are
161: shown in grey (bottom panels). Bins in white correspond to areas not covered by the
162: mobile phone carrier.}
163: \label{fig2}
164: \end{figure*}
165:
166: The call volume shows strong variations with time and day of the week, as shown in Figure 1, but
167: differences across subsequent weeks are generally mild (provided one considers call traffic in the
168: same place, time and day of the week).
169: To capture the weekly periodicity of the observed patterns,
170: we define $n_i({\bf{r}},t,T)$ as the number of calls recorded
171: at location ${\bf{r}}$ (which can either denote a single Voronoi cell or a group of neighboring cells)
172: during the $i$th week between times $t$ and $t+T$, where time is
173: defined modulo 1 week.
174: Assuming we have access to continuous data for $N$ weeks, the mean call activity is given by
175: \begin{equation}
176: \langle n({\bf{r}},t,T)\rangle = {{1}\over{N}}\sum_{i=1}^Nn_i({\bf{r}},t,T)\ .
177: \end{equation}
178: Note that, in the same way as one can trade off spatial resolution for increased
179: statistics by summing over a group of Voronoi cells, varying $T$ one can regulate time accuracy versus
180: statistics. This certainly depends on the extent to which aggregated data shows a regular, stable
181: behavior. The results presented here correspond to $T=1$ hour.
182:
183: \begin{figure}[t]
184: \centerline{{\epsfxsize=5.2in \epsfysize=2.3in \epsfbox{fig3.eps}}}
185: \caption{Size of the largest cluster as a function of the fluctuation
186: threshold for the normal case (left) and the anomalous one (right). Measurements on the call data
187: (solid line with circles) are compared to those of randomized distributions, of which we show
188: the mean (long-dashed line) and confidence bounds at $\pm\sigma_{rdm}$ (short-dashed lines)
189: and $\pm 2\sigma_{rdm}$ (dotted lines).}
190: \label{fig3}
191: \end{figure}
192:
193: The scale to measure
194: departures from the average behavior is set by the {\it standard deviation}, defined as
195: \begin{equation}
196: \sigma({\bf{r}},t,T) = \sqrt{{{1}\over{N-1}}\sum_{i=1}^N
197: \left(n_i({\bf{r}},t,T)-\langle n({\bf{r}},t,T)\rangle\right)^2}\ .
198: \end{equation}
199: Hence, using recorded data for an extended period of time, one can determine the expected call traffic
200: levels and corresponding deviations for all times and locations. Once this {\it normal} behavior is established,
201: {\it anomalous} fluctuations above or below a given threshold can be obtained using the condition
202: \begin{equation}
203: |n_i({\bf{r}},t,T)-\langle n({\bf{r}},t,T)\rangle| > A_{thr}\times\sigma({\bf{r}},t,T)\ ,
204: \end{equation}
205: where $A_{thr}>0$ is a constant that sets the fluctuation level.
206:
207: We grouped Voronoi cells together generating a regular 2D grid made of square bins of
208: about 12 km of linear size. Considering a fixed time slice, we study
209: the spatial clustering of bins showing anomalous activity at different fluctuation levels.
210: In order to illustrate our procedure, Figure 2 shows the activity and fluctuations in a grid of
211: size $40\times 40$ bins (i.e. $480\times 480$ km$^2$ area).
212: We compare the activity in the same region for 2 different weeks (corresponding to the
213: same time and day of the week). The left panels show a {\it normal event}, in which fluctuations
214: around the local mean activity are typically small, with just a few scattered bins having somewhat larger
215: deviations. The right panels, however, show an {\it anomalous event}, characterized by extended, spatially
216: correlated fluctuations that indicate the emergence of a large-scale, coordinated activity pattern. As pointed
217: out above, the existence of anomalous activity patterns could be indicative of possible emergency situations.
218: Similarly to the Voronoi maps already discussed, the upper panels in Fig.2 show the activity (number of calls
219: per hour inside each square bin) in log$_{10}$ scale. White bins correspond to areas not covered by the
220: mobile phone provider. Taking a fixed threshold value $A_{thr}=0.25$,
221: the bottom panels show the high-activity bins above the fluctuation threshold (in black) and the bins with
222: normal activity (in grey). Note that, although the activity maps have a similar appearance to the degree
223: that they seem at first look indistinguishable, the fluctuation maps
224: display striking differences.
225:
226: \begin{figure}[t]
227: \centerline{{\epsfxsize=5.2in \epsfysize=2.3in \epsfbox{fig4.eps}}}
228: \caption{Number of different clusters as a function of the fluctuation
229: threshold for the normal case (left) and the anomalous one (right).
230: Measurements on the call data (solid line with circles)
231: are compared to results on random configurations (dashed and dotted lines).}
232: \label{fig4}
233: \end{figure}
234:
235: In order to quantify the clustering of anomalous bins, we will use the standard tools of percolation theory and
236: determine the size of the largest cluster, the number of
237: different clusters, and the size distribution of all clusters.
238: The statistical significance of the measured clustering is evaluated by comparing it to results from
239: randomized distributions, in which
240: many different configurations are randomly generated, keeping fixed the total number of high-activity
241: bins above the fluctuation threshold. The substrate, which is formed by all bins with non-zero activity,
242: remains always the same (in Fig.2, for instance, the substrate is the set of all grey and black bins). Clusters
243: are defined by first- and second-order nearest neighbors in the square 2D grid.
244: In the remainder of this section, we will focus on a specific large-scale anomalous event and
245: compare it to the normal behavior
246: observed in data of a different week (but corresponding to the same time and day of the week). The comparison
247: between normal and anomalous events will illustrate the use of percolation observables as diagnostic tools
248: for anomaly detection.
249:
250: \begin{figure}[t]
251: \centerline{{\epsfxsize=5.2in \epsfysize=3.2in \epsfbox{fig5.eps}}}
252: \caption{Cumulative size distribution of all clusters as a function of cluster
253: size, for $A_{thr}=0.25$ (upper panels), $A_{thr}=0.75$ (bottom panels), normal case
254: (left panels), and anomalous case (right panels). Thick solid lines are measurements on the call data,
255: while dashed and dotted lines are results from random configurations.}
256: \label{fig5}
257: \end{figure}
258:
259: Figure 3 shows the size of the largest cluster, $S_{max}$, as a function of the fluctuation
260: threshold $A_{thr}$, for the
261: normal case (left) and the anomalous one (right).
262: Each measured plot (solid line with circles) is compared to results from randomized
263: distributions. The latter correspond to the mean (long-dashed line)
264: and confidence bounds at $\pm\sigma_{rdm}$ (short-dashed lines)
265: and $\pm 2\sigma_{rdm}$ (dotted lines), as obtained from generating 100 random configurations in each case.
266: As expected, the plots show that the size of the largest cluster monotonically decreases with the fluctuation threshold.
267: However, while the clustering in the normal case lacks any significance, the anomalous event shows large departures
268: from the clustering expected in a random configuration.
269:
270: In the same vein, Figure 4 shows the number of different clusters, $N_{cl}$, as a
271: function of the fluctuation threshold $A_{thr}$, where measurements on the call data for the same
272: normal (left) and anomalous (right) events are compared to results from randomized configurations.
273: As before, in the normal case the number of clusters agrees well with the expectations for random configurations,
274: while significant departures are observed in the anomalous case.
275:
276: Figure 5 shows the cumulative size distribution of all clusters, $N_{cl}(s_{cl}>S)$, as a function of the cluster
277: size $S$, compared to random configurations. The upper panels display results for $A_{thr}=0.25$, while
278: the bottom ones show results for $A_{thr}=0.75$, as indicated.
279: Moreover, the left panels correspond to the normal event,
280: while the right panels to the anomalous event. Again, the measured cluster size distribution in the normal
281: case is in good agreement with the expected one for a random configuration. In contrast, the anomalous event shows the
282: occurrence of a few very large clusters formed by many highly active bins.
283: These unusually large structures cannot be explained as arising just from random configurations, but instead are
284: the result of the spatiotemporal correlation of large, highly active regions.
285:
286: As a summary, in this Section we showed how large-scale collective behavior
287: can be described using aggregated data resolved in both time and space.
288: Moreover, we developed the basic framework
289: for detecting and characterizing spatiotemporal fluctuation patterns,
290: which is based on standard procedures of statistics and percolation theory.
291: These tools are particularly effective in detecting extended anomalous events,
292: as those expected to occur in emergency scenarios due to e.g. natural
293: catastrophes and terrorist attacks.
294:
295: \section{Individual calling activity patterns}
296: In order to use the huge amount of data recorded by
297: mobile phone carriers to investigate
298: various aspects of human
299: dynamics~\cite{gon07,laszlo,list41,list42,list43},
300: a necessary starting point it is to characterize
301: the dynamics of the individual calling activity {\it per se}.
302: Previous studies have measured the time between consecutive
303: individual-driven events, such as sending e-mails, printing,
304: and visiting web pages or the library~\cite{Olivera,Maya}.
305: Those events are described by heavy-tailed
306: processes~\cite{laszlo,Goh}, challenging the traditional Poissonian modeling
307: framework~\cite{VazquezPRL1,Caldarelli,Blanchard,Daly,Cesifoti}, with consequences on
308: task completion in computer systems.
309: In this section we explore the interevent distribution
310: of the calling activity of $6 \times 10^{6}$ mobile phone users during $1$
311: month.
312:
313: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
314: \begin{figure*}[t]
315: \begin{center}
316: \includegraphics*[width=10.0cm]{fig6.eps}
317: \end{center}
318: \caption{\protect Interevent time distribution
319: $P(\Delta T)$ for calling activity.
320: $\Delta T$ corresponds to the time interval
321: between two mobile phone calls sent by the same user.
322: Different symbols indicate the measurements done over
323: groups of users with different activity levels (\# calls).
324: The inset shows the unscaled interevent time
325: distribution and the solid line corresponds to
326: Eq.~(\ref{eq:distr}).}
327: \label{fig6}
328: \end{figure*}
329: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
330:
331: As many other human activities, the calling activity
332: pattern is highly heterogeneous.
333: While some users rarely use the mobile phone, others
334: make hundreds or even thousands of calls each month.
335: To analyze such different levels of activity, we group
336: the users based on their total number of calls.
337: Within each group, we measure the probability density
338: function $P(\Delta T)$ of the time interval $\Delta T$ between
339: two consecutive calls made by each user.
340: As shown by the inset of Fig.~\ref{fig6}, the tail of the distribution
341: is shifted to longer interevent times for users with less activity.
342: However, if we plot $\Delta T_{a} P(\Delta T)$ as a function
343: of $\Delta T/\Delta T_{a}$, where $\Delta T_{a}$ is
344: the average interevent time for the corresponding user, the data collapses into a
345: single curve (Fig.~\ref{fig6}). This
346: indicates that the measured interevent distribution follows the expression
347: $P(\Delta T)$ = $1/\Delta T_{a}$$\mathcal{F}$$(\Delta T / \Delta T_{a})$, where
348: $\mathcal{F}$$(x)$ is independent from the average
349: activity level of the population. This represents a
350: universal characteristic of the system that surprinsingly also
351: coincides with results from e-mail communication~\cite{cond-matGoh}.
352: The data are well fitted by
353: \begin{equation}
354: P(\Delta T) = (\Delta T)^{-\alpha} \exp (\Delta T/\tau_{c}),
355: \label{eq:distr}
356: \end{equation}
357: where the power law scaling with exponent $\alpha = 0.9 \pm 0.1$ is followed
358: by an exponential cutoff at $\tau_{c} \approx 48$ days.
359: Equation (\ref{eq:distr}) is shown by a solid line in the inset
360: of Fig.~\ref{fig6} and its scaled version is presented
361: in the main panel of the figure using $\Delta T_{a}=8.2$ hours,
362: which is the average interevent time measured for the whole population.
363: This result, clearly different from the one predicted
364: by a Poisson approximation~\cite{Goh,Feller,Sornette}, would for
365: instance affect the predictions of spreading dynamics
366: through the network of calls~\cite{VazquezPRL2}.
367:
368: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
369: \begin{figure*}[t]
370: \begin{center}
371: \includegraphics*[width=10.0cm]{fig7.eps}
372: \end{center}
373: \caption{\protect Travel behavior. {\bf (a)-(b)} Number of trips
374: and consecutive calls that are reported within a fixed interevent
375: time $\Delta T_{o}=30$ min vs. time of the day.
376: {\bf (c)} The ratio of the two quantities described
377: in (a) and (b) shows that along the whole day $40 \pm 20 \%$
378: of the people that is calling seems to be also traveling.
379: {\bf (d)} The average distance of travel within $\Delta T_{o}=30$ min
380: remains constant during the day within $6 \pm 2$ km,
381: a reasonable value that may correspond to the combination between
382: walk and motor transportation.}
383: \label{fig7}
384: \end{figure*}
385: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
386:
387:
388: To explore the interplay between human activity and
389: mobility patterns, we fix the characteristic observation
390: time to $\Delta T_{o} = 30$ min and collect only those
391: consecutive calls that occur with this
392: interevent time, recording also the time of the day
393: in which they occurred (Fig. \ref{fig7} a).
394: For each pair of calls, we count how many of them
395: result in a change of coordinate, e.g. the user
396: traveled in the $30$ min time interval between the calls
397: (Fig. \ref{fig7} b). The number of events that result in a change
398: of location and the number of calls as a function of time capture the daily
399: activity pattern of the users~\cite{Huberman}.
400: We find that both the call and the mobility pattern
401: decrease at night and have clear peaks near noon and late
402: evening. There is a factor of $30$ between the largest and the smallest
403: number of events (calls/changes of location)
404: reported during the day. Interestingly, when
405: we calculate the fraction of consecutive calls also resulting
406: in a potential change of location, the quantity varies
407: at most $40 \%$ during the whole
408: day (Fig.~\ref{fig7}c). This indicates that although the total activity
409: varies strongly, the percentage of the people
410: that are calling and traveling remains rather stable.
411: More importantly, the average distance traveled within $\Delta T_{o} = 30$
412: min. is stable in the vicinity of $\Delta r = 6 \pm 2$ km
413: (Fig.~\ref{fig7}d), a value consistent for the combination between
414: walk and motor transportation.
415:
416: \section{Conclusions}
417:
418: Novel aspects of human dynamics and social interactions were addressed by means of
419: mobile phone data with time and space resolution.
420: This allowed us to study the mean collective behavior at
421: large scales and focus on the occurrence of anomalous events.
422: %The basic spatial unit, a Voronoi cell defined by the
423: %distribution of transceiver antennas,
424: Considering a fixed time slice,
425: we partitioned the space using a regular grid and studied the aggregated call activity inside each
426: square bin forming the grid.
427: We showed that anomalous events give rise to spatially extended patterns that can
428: be meaningfully quantified in terms
429: of standard percolation observables.
430: By considering a series of consecutive time slices, we could investigate the
431: rise, clustering and decay of spatially extended anomalous events, which could be
432: relevant e.g. in real-time detection of emergency situations.
433:
434: We also investigated patterns of calling activity at the individual level.
435: We observed that
436: the interevent time of consecutive calls is heavy-tailed,
437: a finding that has implications for dynamics of spreading phenomena on social networks, and that
438: agrees with results previously reported on other, related human activities.
439: We also show that, despite of the complexity inherent in the
440: interevent calling patterns,
441: it is still possible to recover some
442: characteristic values from the behavior
443: of the population that are stationary during the day,
444: such as the fraction of active traveling population
445: and their average distance traveled.
446:
447: In many ways, these results represent only a first step towards understanding human activity patterns.
448: Our results indicate that the rich information provided by mobile communication data
449: open avenues to addressing novel problems. These tools offer a chance to improve our
450: understanding of complex networks as well \cite{CNP1,CNP2,CNP3,CNP4,CNP5,CNP6,CNP7,CNP8},
451: by potentially correlating the structure of social
452: networks with the spatial layout of the users as nodes \cite{SNP1,SNP2,SNP3,SNP4,gon_prl,lind1,lind2},
453: thus contributing to a better understanding
454: of the spatiotemporal features of network evolution.
455:
456: \section*{Acknowledgments}
457: This work was supported by the James S. McDonnell Foundation 21st Century Initiative in
458: Studying Complex Systems, the NSF within the DDDAS (CNS-0540348),
459: ITR (DMR-0426737) and IIS-0513650 programs, as well as by U.S. Office of Naval Research N00014-07-C
460: and the NAP Project sponsored by the National Office for Research and Technology (KCKHA005).
461: Data analysis was performed on the Notre Dame Biocomplexity Cluster
462: supported in part by NSF MRI Grant No. DBI-0420980.
463:
464: \begin{thebibliography}{99}
465: \bibitem{gon07} M.C. Gonz\'alez and A.-L. Barab\'asi, Nature Phys. {\bf 3}, 224 (2007).
466: \bibitem{onn07a} J.-P. Onnela, J. Saram\"aki, J. Hyv\"onen, G. Szab\'o, D. Lazer, K. Kaski,
467: J. Kert\'esz, and A.-L. Barab\'asi, Proc. Nat. Acad. Sci. {\bf 104}, 7332 (2007).
468: \bibitem{onn07b} J.-P. Onnela, J. Saram\"aki, J. Hyv\"onen, G. Szab\'o, M. A. de Menezes,
469: K. Kaski, A.-L. Barab\'asi, and J. Kert\'esz, New J. Phys. {\bf 9}, 179 (2007).
470: \bibitem{sza06} G. Szab\'o and A.-L. Barab\'asi, arXiv:physics/0611177.
471: \bibitem{ratti1} C. Ratti, R.M. Pulselli, S. Williams, and D. Frenchman,
472: Environment and Planning B {\bf 33}, 727 (2006).
473: \bibitem{ratti2} C. Ratti, A. Sevtsuk, S. Huang, and R. Pailer, {\it Location Based Services and TeleCartography}
474: (Springer, Berlin, Heidelberg, 2007), Sect. V, p. 433.
475: \bibitem{pal07a} G. Palla, A.-L. Barab\'asi, and T. Vicsek, Nature {\bf 446}, 664 (2007).
476: \bibitem{pal07b} G. Palla, A.-L. Barab\'asi, and T. Vicsek, Fluct. Noise Lett. {\bf 7}, L273 (2007).
477: \bibitem{pas01} R. Pastor-Satorras and A. Vespignani, Phys. Rev. Lett. {\bf 86}, 3200 (2001).
478: \bibitem{Zoltan} S. Eubank, H. Guclu, V.S.A. Kumar, M. Marathe,
479: A. Srinivasan, Z. Toroczkai, and N. Wang,
480: Nature {\bf 429}, 180 (2004).
481: \bibitem{Grenfell:Science} C. Vibpoud, O. Bjonstadt, D.L. Smith, L. Simonsen,
482: M. A. Miller, and B.T. Grenfell, Science {\bf 312}, 447 (2006).
483: \bibitem{Vespignani} V. Colizza, A. Barrat, M. Barthelemy, A.-J. Valleron, and A.Vespignani,
484: PLoS Medicine 4(1): e13 (2007).
485: \bibitem{gon_epi1} M.C. Gonz\'{a}lez and H.J Herrmann,
486: %"Scaling of the propagation of epidemics in a system of mobile agents",
487: Physica A {\bf 340}, 741 (2004).
488:
489: \bibitem{gon_epi2} M.C. Gonz\'{a}lez, H.J. Herrmann, and A.D. Ara\'{u}jo,
490: %"Cluster size distribution of infection in a system of mobile agents",
491: Physica A, {\bf 356}, 100 (2005).
492:
493: \bibitem{can06} J. Candia, Phys. Rev. E {\bf 74}, 031101 (2006).
494: \bibitem{can07a} J. Candia, Phys. Rev. E {\bf 75}, 026110 (2007).
495: \bibitem{can07b} J. Candia, J. Stat. Mech. P09001 (2007).
496:
497: \bibitem{mad06} G. Madey, G. Szab\'o, and A.-L. Barab\'asi, in {\it Lecture Notes in
498: Computer Science}, V.N. Alexandrov, G.D. van Albada, P.M.A. Sloot, and J. Dongarra (Eds.),
499: (Springer, Berlin, 2006) Vol. 3993, p. 417.
500: \bibitem{sch07} T. Schoenharl, R. Bravo, and G. Madey, Int. J. Intel. Contr. Sys. {\bf 11}, 209 (2007).
501: %\bibitem{sta94} D. Stauffer and A. Aharony, {\it Introduction to Percolation Theory}
502: %(2nd Ed.), (Taylor and francis, London, 1994).
503: \bibitem{laszlo} A.-L. Barab\'{a}si, Nature {\bf 435}, 207-211, (2005).
504:
505: \bibitem{list41} A. V\'{a}zquez,
506: %Impact of memory on human dynamics
507: Physica A {\bf 373}, 747 (2007).
508:
509: \bibitem{list42} Z. Dezs\"{o}, E. Almaas, A. Luk\'acs, B. R\'acz, I. Szakad\'at,
510: and A.-L. Barab\'{a}si,
511: %Dynamics of information access on the web
512: Phys. Rev. E {\bf 73}, 066132 (2006).
513:
514: \bibitem{list43} D. Helbing, M. Treiber, and A. Kesting,
515: %Understanding interarrival and interdeparture time statistics from
516: %interactions in queuing systems
517: Physica A {\bf 363}, 62 (2006).
518:
519:
520: \bibitem{Olivera}J. G. Oliveira and A.-L. Barab\'{a}si, Nature {\bf 437}, 1251 (2005).
521:
522: \bibitem{Maya} U. Harder and M. Paczuski, Physica A {\bf 361}, 329 (2006).
523:
524: \bibitem{Goh} A. V\'{a}zquez, J. G. Oliveira, Z. Dezs\"{o}, K.-I. Goh,
525: I. Kondor, and A.-L. Barab\'{a}si,
526: %Modeling bursts and heavy tails in human dynamics
527: Phys. Rev. E {\bf 73}, 036127 (2006).
528:
529: \bibitem{VazquezPRL1} A. V\'{a}zquez,
530: %Exact Results for the Barabási Model of Human Dynamics
531: Phys. Rev. Lett. {\bf 95}, 248701 (2005).
532:
533: \bibitem{Caldarelli} A. Gabrielli and G. Caldarelli,
534: %Invasion percolation and critical transient in the Barabasi model of human dynamics
535: Phys. Rev. Lett. {\bf 98}, 20 (2007).
536:
537: \bibitem{Blanchard} P. Blanchard and M.O. Hongler,
538: %Modeling human activity in the spirit of Barabasi's queueing systems
539: Phys. Rev. E {\bf 75}, 026102 (2007).
540:
541: \bibitem{Daly} E. Daly and A. Porporato,
542: %Intertime jump statistics of state-dependent Poisson processes
543: Phys. Rev. E {\bf 75}, 011119 (2007).
544:
545: \bibitem{Cesifoti} C. Hidalgo, Physica A {\bf 369}, 877 (2006).
546:
547: \bibitem{cond-matGoh} K.-I. Goh and A.L. Barab\'{a}si,
548: %Burstiness and memory in complex systems.
549: arXiv:physics/0610233.
550:
551: \bibitem{Feller} W. Feller, {\it An Introduction to Probability Theory and its
552: Applications} (Wiley, New York, 1966), Vol. II.
553:
554: \bibitem{Sornette} J. Laherr\`{e}re and D. Sornette,
555: %"Stretched exponentials distributions in nature and economy: fat tails
556: %with characteristics scales."
557: Eur. Phys. J. B {\bf 2}, 525 (1998).
558:
559: \bibitem{VazquezPRL2} A. V\'{a}zquez, B. R\'acz, A. Luk\'acs, and
560: A.-L. Barab\'{a}si,
561: %Impact of non-Poisson activity patterns on spreading processes,
562: Phys. Rev. Lett. {\bf 98}, 158702 (2007).
563:
564: \bibitem{Huberman} S. A. Golder, D. Wilkinson, and B. A. Huberman,
565: %"Rhythms of Social Interaction: Messaging within a Massive Online Network"
566: 3rd Int. Conf. on Communities and Technologies (CT2007).
567: East Lansing, MI. June 28-30, (2007).
568:
569: \bibitem{CNP1} A.-L. Barab\'asi and R. Albert, Science {\bf 286}, 509 (1999).
570: \bibitem{CNP2} R. Albert and A.-L. Barab\'asi, Rev. Mod. Phys. {\bf 74}, 47 (2002).
571: \bibitem{CNP3} S.N. Dorogovtsev and J.F.F. Mendes, {\it Evolution of networks: From Biological
572: Nets to the Internet and WWW} (Oxford University Press, Oxford, 2003).
573: \bibitem{CNP4} R. Pastor-Satorras and A. Vespignani, {\it Evolution and structure of the Internet}
574: (Cambridge University Press, Cambridge, 2004).
575: \bibitem{CNP5} S. Boccaletti, V. Latora, Y. Moreno, M. Chavez, and D.-U. Hwang, Phys. Rep. {\bf 424}, 175 (2006).
576: \bibitem{CNP6} M. Newman, A.-L. Barab\'asi, and D.J. Watts, {\it The Structure and Dynamics of Networks}
577: (Princeton University Press, Princeton and Oxford, 2006).
578: \bibitem{CNP7} G. Caldarelli, {\it Scale-Free Networks} (Oxford University Press, Oxford, 2007).
579: \bibitem{CNP8} G. Caldarelli and A. Vespignani (Eds.), {\it Large Scale Structure and Dynamics
580: of Complex Networks} (World Scientific, Singapore, 2007).
581: \bibitem{SNP1} S.S. Manna and P. Sen, Phys. Rev. E {\bf 66}, 066114 (2002).
582: \bibitem{SNP2} S. H. Yook, H. Jeong, and A.-L. Barab\'asi, Proc. Natl. Acad. Sci. USA {\bf 99}, 13382 (2003).
583: \bibitem{SNP3} A. Barrat, M. Barth\'el\'emy, and A. Vespignani, J. Stat. Mech. P05003 (2005).
584: \bibitem{SNP4} G. Grinstein and R. Linsker, Phys. Rev. Lett. {\bf 97}, 130201 (2006).
585:
586: \bibitem{gon_prl} M.C. Gonz\'{a}lez, P.G. Lind, and H.J. Herrmann,
587: Phys. Rev. Lett. {\bf 96}, 088702 (2006).
588: \bibitem{lind1} P.G. Lind, J.S. Andrade Jr., L.R. da Silva, and H.J. Herrmann,
589: Phys. Rev. E {\bf 76}, 036117 (2007).
590: \bibitem{lind2} P.G. Lind, J.S. Andrade Jr., L.R. da Silva, and H.J. Herrmann,
591: Europhys. Lett. {\bf 78}, 68005 (2007).
592:
593: \end{thebibliography}
594: \end{document}