Social networks as economic and social sensors

Temporal, structural and activity patterns in social networks are related to human behavior. Thus, networks do have different shape and dynamics under different exogenous and endogenous shocks or conditions. In the last years we have addressed the question of whether we can use our understanding of social networks to anticipate, predict or measure important phenomena as information propagation, disaster damage, unemployment shocks, weather conditions, gender digital divide, etc.

For example in a series of papers, we demonstrated for the first time the ability of the so-called “friendship paradox” in social networks to get better collection of users (sensors) that can anticipate meme, news or event-related information propagation in networks. Even those sensors can be user to disaster management as we found during Hurricane Sandy.

We also found that social media behaviors like mobility around economical areas (see figure), diversity of interactions between areas or the gramatical correctness of posts are deeply related to unemployment in those areas. Those findings have been corroborated in other countries, showing the universal character of those fingerprints of socio-economic status in social media. In the last years we have also used data from social media to understand gender inequality. By developing a unique large-scale dataset from Facebook, we have seen the relation between the activity of 1.4 billion users in 217 countries, the digital gender gap and other inequality measures related to education, health or economic opportunity. This research was published in PNAS \cite{Garcia:2018bv} and together with the methodology in previous papers we have use these results in a collaboration with UNICEF to measure Human Development Indexes in third world countries, where official data is scarce or incomplete.

Some recent papers:

Talks:

• From individual behavior to social processes: using human traces for social good Invited talk at ICCS 2018 (Boston).

Videos:

Media: