Personal Space and Automatically Learned Social Networks
0Even networks of very simple sensors, when the data is subjected to sufficient analysis, can reveal deep and subtle information about the behaviors of people moving around them. Social networks can be automatically learned and updated by applying machine learning to sensor network data. This talk describes a social network that is automatically learned from a sensor network which has been installed in an office environment. Describing some recent work conducted in collaboration with colleagues at MERL, this talk will cover two aspects of the social networking system. Firstly, I'll compare its performance relative to self-reported social connections. Secondly, I'll analyze questionnaire feedback regarding the privacy implications of automatically-generated social networks to understand how people perceive these new technologies. The talk will show that the automatic social network learning system was able to estimate 86 false positives. The questionnaire was distributed to users familiar with the system and Internet respondents who were unfamiliar with the system. The research found that these respondents feel their privacy is invaded by many possible methods of gathering social network information. Those who had used the automatic social network learning system, however found that motion sensor logs were less invasive than other possible observation channels.