Complexity and Social Networks Blog

« eGooglement - How google is trying to improve the accesibility of government websites | Main | Wayne Baker on Social Capital and Generalized Reciprocity »

3 May 2007

Automatically Sensing Social Networks

My group at the MIT Media Lab has started conducting experiments using the Sociometric Badge, a sensing platform that logs voice features, proximity to other individuals, face-to-face interactions, and movement. We have finished analysis of data obtained in a preliminary study at a German bank’s marketing division, and the results were astonishing. We were also fortunate enough to get the e-mail logs for the bank over the course of the study.

For years e-mail data has been used as an easy to obtain proxy for social network information. We found, however, that in fact proximity is highly negatively correlated with e-mail use. We can attribute this to several factors. First, if you are in close proximity to another individual, it makes more sense to interact with them in the real world rather than send them an e-mail. Second, proximity information also picks up on informal relations, while in this particular organization e-mail is used mainly for business purposes. This is because if you spend a lot of time with someone you are more likely to be their friend and therefore less likely to send an e-mail to them. This result points towards the necessity of having face-to-face interaction information in order to have a full view of the social network.

When we combined face-to-face interaction information (which we used to create a social network representation) with e-mail data we got a more complete view of the social network that exhibited some interesting properties. Most notably, betweenness in the social network and total communication were both highly negatively correlated with the perceived quality of interactions. These two measures, however, were not significantly correlated with each other. This points to the necessity of aiding central actors in managing their interaction-related stress, since it is evident that those who are overburdened with their communication responsibilities feel that their interaction quality similarly degrades.

These results strongly argue for the use of automatic sensing data for computational social science. We report many other interesting findings in a paper that will appear in the NetSci 2007 conference.

Posted by Ben Waber at May 3, 2007 10:18 AM

.
• Business (3)
.
• Citizen Relationship Management (17)
.
• Commentary (7)
.
• Complexity (2)
.
• Current Trends (47)
.
• eGovernment (21)
.
• Events/Announcements (72)
.
• Innovation (14)
.
• Internet (18)
.
• Knowledge Sharing (20)
.
• Methodology (32)
.
• Networked Governance (18)
.
• Networks in political science (5)
.
• PNG Working Paper Series (6)
.
• Powerlaws (4)
.
• Reviews (2)
.
• Simulations (1)
.
• Small group networks (3)
.
• Small Worlds (2)
.
• SN in the news (72)
.
• Social Capital (3)
.
• Social Computing (12)
.
• Social Finance (1)
.
• Social Networking Platforms (11)
.
• Social Psychology (2)
.
• Social Software (25)
.
• Societal Networks (40)
.
• Technology (35)