Finding Social Behavior Patterns Through Call Detail Records
How Mobile Operator Data Can Give us Insights about Social Segregation?
Social behavior is related to mobility patterns in the sense that social and other external factors can influence, for example, our choice of transportation mode or our route choices. Yuan et al. [1] correlate some mobility measures with the gender and the age of the users. Among other interesting conclusions, it was discovered that adolescents and the elderly do not travel as long distances as the middle-aged and young people. Isaacman et al. [2] concluded that people tend to travel more in the summer than in the winter. So, the seasons can actually influence our amount of travel. Furthermore, Cho et al. [3] show us that long-distance trips are way much influenced by our social ties when compared to short-distance ones. They used CDRs (Call Detail Records) to characterize the quantity, direction, duration, and seasonality of the migration. Blumenstock [4] was also able to infer internal migration patterns through the analysis of CDRs.
According to Deville et al. [5], CDRs are, in fact, a great tool to predict population movements. The contributions of Wang et al. [6] and Tatem et al. [7] are relevant to understand the epidemiology phenomenon and virus propagations by characterizing human mobility through the mobile operator data. Also, Eagle et al. [8] help us understand how people adapt and change their behavior in communication to be more similar to their new social environment. The discovers from this study constitute a critical source of information to event management and congestion reduction.
With the purpose of examining the evolution of the tie strength, sociality levels and other factors among users’ social ties during migration periods, Phithakkitnukoon et al. [9] made use of 11 months of CDRs from Portugal. Valuable conclusions were discovered by Krings et al. [10] as well. For example, the following formula to characterize the intensity of the communications between two cities was obtained: the intensity is “proportional to the product of the two populations divided by the square of the distance between the cities.”. Besides that, it was observed that ”intra-urban communications scale superlinearly with city population” [10].
In this line of reasoning, it is opportune to talk about discrete-choice models. These models allow us to find the probability that the user uses determined transport modes through a function that has into account multiple factors [11]. So, this model assumes that the user’s behaviors are dictated by the maximum gain possibly obtained and the attractiveness of other competitors' alternatives. However, our choices are not so rational as we might think and is by having that in mind that Phithakkitnukoon et al. [11] try to establish a relationship between sociability measures and user’s mobility patterns. Then, the concept of homophily is central in the sense that we need to be aware that we tend to socialize and form connections with people that are similar to us. These people tend to share with us common characteristics or possessions (e.g., gender or age). So, it is just rational to extrapolate that the more the number of our social ties that use a particular travel mode, the bigger the likelihood of us using that same travel mode. The results of this study proved that our most closer ties have a stronger influence in choosing private transportation. Reversely, our weakest relationships are those that persuade us more to adopt public transports. Besides that, it is also curious that friends that are geographically closer to us have more power in our choice of transportation in our commuting trips. As expected, it was also found that the distance to access public transports contributes to reject it.
Looking closely into the work of Phithakkitnukoon et al. [12], we see that it is possible to extract mobility measures like mobility diversity, mobility dispersion, and range from CDRs. Mobility diversity is the “total number of different locations visited” by the user. Mobility dispersion “measures the amount of variation (or randomness) in mobility” [12]. Mobility range “infers the travel distance range of the person’s mobility, which is defined as the distance (in kilometers) from the person’s home location to the farthest location the person ever visited” [12]. Besides mobility measures, sociability measures also were extracted from CDRs (e.g., call frequency, call duration, and the number of social ties). All these indicators are analyzed in order to know which one of them influences our mobility patterns, namely, our choice of transportation mode. Through CDRs, we can calculate how intense are our social interactions and, therefore, infer some important shared characteristics between us and our most strong social ties. However, as this kind of data is always anonymized, we cannot know much more about the user beyond his/her location and the users to whom he/she is calling.
After the comparison between the six measurements of each user and his/her social relationship, it was concluded that mobility diversity is proportional to the strength of the social ties. It is proportional in the way that, the stronger the social connection, the more similar it is their mobility diversity and mobility range. Conversely, it was found that mobility dispersion is not correlated at all with the strength of social ties. In what concerns to sociality measures, it was discovered that all the three measurements are similar between the user and their closest social relations. So, as we tend to have similar social behavior to our closest friends, this research shows us that it is possible to infer some mobility patterns from the behaviors observed in our closest social connections. Mobility dispersion is the only indicator that deviates from this conclusion. With this study, homophily philosophy was enforced. The underlying problem in this study is that it is constrained to analyze only the social ties that share the same mobile operator.
It is becoming crucial to know how to calculate the social strength among social ties. For that, we need to recall the work of Granovetter et al. [13]. They defined it as the “combination of the amount of time, the emotional intensity, the intimacy (mutual confiding), and the reciprocal services”. So, inspired by the work of Nicholas et al. [14], in our case, we can associate the amount of time of communicating through voice calls to the sociability measurement that fits the Granovetter previous definition. To do so, we need to consider the following ratio:
That formula means that the user social strength (s) with social tie i is equal to the amount of time that the user spent talking with tie i — c(i) — divided by the amount of time that the user spent talking with all their social ties (N). In order to compute this calculation correctly, it was fundamental to consider social connections as those who maintain reciprocal calls with the observed user. We say it is vital because there are often calls that we make sporadically to some service or to someone that is not for sure our friend or acquaintance (to handle some business, for example).
In what infer a social network concerns, CDRs may be positioned as one of the best opportunistic data to use. That is a reasonable conclusion once the people to/from which we make/receive calls mimic the closest as possible our real social network. For example, if we took the data from social networks like Facebook or Instagram and based our social ties on the so-called “friends” or “followers” in those platforms, we would end up overestimating by far the actual number of user’s social ties.
Olivier et al. [15] also investigate the effects of weather conditions on our social interactions. The weather has an impact on socio-economic activities. People may prefer to reside in a determined zone because of the weather conditions around them. Besides conditioning social interactions, changes in weather conditions also influence mobility patterns. Along with CDRs, a weather dataset was needed for this investigation. The dataset came from three base stations in Lisbon that measured temperature, humidity, and pressure every 30 minutes. In the end, it was concluded that weather conditions do not have a significant influence on people’s average talk time. In extreme temperatures (excessively cold or warm) and pressures (high and low) as well as humidity levels between 20%-100%, people are willing to talk to a smaller number of social ties and maintain more connections with their strong relationships.
Despite having valuable conclusions, this research has some issues. The influence of the atmospheric parameters might depend on each other; however, only the importance of one criterion at a time was explored. It could have been investigated the impact of the simultaneous changes in multiple parameters of the weather conditions. Also, the study ignores the fact that people might be inside some infrastructure and be shielded from certain types of weather. It was neither considered the weather conditions of the person who was connected to the observed individual. The fact that there are holidays, vacations, and other social events that can influence the average talk time was not taken into consideration as well.
The research developed by Olivier et al. [16] constitutes another great study about the socio-geography of human mobility. In this study was discovered that 80% of the locations visited are within a range of 20 km of the nearest users’ social ties. If we consider a geo-social radius of 45 km, then we can say that the percentage increases to 90%. It is also surprising to know that we tend to be geographically closer to our weak ties than our strong ties. In general, the more urbanely dense is our region, the more distant we are from our social ties and the shorter are our geo-social radius; also, 80% of our travel scope will be within a 10 km geo-social radius. Here, the geo-social radius is defined as being the geographical distance from social ties. So, if the “places visited by the subject (or travel scope) are within x km from the subject’s social ties”, then the geo-social radius will be x. This concept is appropriately depicted in Figure 1.
A more general and surprising conclusion of the study is that “although people tend to reside near their strong ties, their mobility is biased towards the geographic locations of their weak ties” [16]. This study, like the others, raises some issues. We are assuming that we make/receive phone calls to every person that is our friend in the period under analysis, which is not realistic. Also, we all have social ties to whom we speak on a regular basis, face-to-face, and we do not need to call them. It was also assumed that people did not migrate or did not change the homeplace, moving to another location during the period of the study. People may also be residing temporarily in someplace because they are on vacation. However, as vacation is just a short slice of people’s lives, it is quite unlikely that it has a significant impact on the study.
References
[1] Y. Yuan, M. Raubal, and Y. Liu, “Computers , Environment and Urban Systems Correlating mobile phone usage and travel behavior — A case study of Harbin , China,” Comput. Environ. Urban Syst., vol. 36, no. 2, pp. 118–130, 2012.
[2] S. Isaacman, R. Becker, S. Kobourov, M. Martonosi, J. Rowland, and A. Varshavsky, “Ranges of Human Mobility in Los Angeles and New York,” pp. 1–6.
[3] E. Cho, “Friendship and Mobility : User Movement In Location-Based Social Networks.”
[4] J. Blumenstock, “Inferring Patterns of Internal Migration from Mobile Phone Call Records : Evidence from Rwanda,” 2012.
[5] P. Deville, C. Linard, S. Martin, M. Gilbert, F. R. Stevens, and A. E. Gaughan, “Dynamic population mapping using mobile phone data,” vol. 111, no. 45, 2014.
[6] P. Wang, M. C. González, C. A. Hidalgo, and A. Barabási, “Understanding the spreading patterns of mobile phone viruses,” vol. 1076, pp. 1071–1076, 2009.
[7] A. Tatem and D. L. Smith, “Quantifying the Impact of Human Mobility on Malaria Quantifying the Impact of Human Mobility on Malaria,” no. March 2014, 2012.
[8] N. Eagle & M. A. Bettencourt, “Community Computing : Comparisons between Rural and Urban Societies using Mobile Phone Data,” pp. 144–150, 2009.
[9] F. Calabrese, “Out of Sight Out of Mind — How Our Mobile Social Network Changes during Migration,” 2018.
[10] C. Link, G. Krings, F. Calabrese, C. Ratti, and V. D. Blondel, “Scaling Behaviors in the Communication Network Between Cities,” 2012.
[11] S. Phithakkitnukoon, T. Sukhvibul, M. Demissie, Z. Smoreda, J. Natwichai, and C. Bento, “Inferring social influence in transport mode choice using mobile phone data,” EPJ Data Sci., vol. 6, no. 1, 2017.
[12] S. Phithakkitnukoon and Z. Smoreda, “Influence of social relations on human mobility and sociality: a study of social ties in a cellular network,” Soc. Netw. Anal. Min., vol. 6, no. 1, pp. 1–9, 2016.
[13] M. S. Granovetter, T. American, and N. May, “The Strength of Weak Ties,” vol. 78, no. 6, pp. 1360–1380, 2007.
[14] A. Nicholas, J. Onnela, S. Arbesman, and M. C. Gonza, “Geographic Constraints on Social Network Groups,” vol. 6, no. 4, 2011.
[15] S. Phithakkitnukoon, T. W. Leong, Z. Smoreda, and P. Olivier, “Weather Effects on Mobile Social Interactions: A Case Study of Mobile Phone Users in Lisbon, Portugal,” PLoS One, vol. 7, no. 10, pp. 1–13, 2012.
[16] S. Phithakkitnukoon, Z. Smoreda, and P. Olivier, “Socio-geography of human mobility: A study using longitudinal mobile phone data,” PLoS One, vol. 7, no. 6, pp. 1–9, 2012.
