March 21, 2015 March 23, 2015

AAAI Spring Symp. on Observational Studies through Social Media and Other Human-Generated Content

Location: Stanford University

Postscript: Thanks to everyone who joined us in March for the engaging, thought-provoking and lively OSSM symposium. And a special thanks to all our speakers and authors, our program committee and also to the organizers of the 2016 AAAI spring symposium series.

More information about the AAAI 2016 Spring Symposium Series.

While using the Internet and mobile devices, people create data, whether intentionally or unintentionally, through their interaction with messaging services, websites and other applications and devices. This means that experiments with heretofore unprecedented populations can be performed in a variety of topics. Our workshop will focus on observational studies which arise from these interactions and data, with a focus on experiments that can indicate causal inferences.

Human generated content in general, and social media in particular, are a rich repository of data for observational studies across many areas: public health, with research on prevalence of disease and on the effects of media on the development of disease; medicine, showing the ability to detect mental disease in individuals using social media; education, to optimize teaching and exams; and sociology, to prove theories previously tested on very small populations. These studies were conducted from data including social media, search engine logs, location traces, and other forms of human generated content.

While many past studies showed a correlation between variables of interest, some studies were able to show causal relationships through natural experiments or by linking data sources. Our workshop focuses on all aspects of causal inference from human generated content, with studies that developed novel methods of identifying and using natural experiments or other methods for inferring causality.

Topics include:

  • Interpreting user-generated data, including text, structured data, and temporal data.
  • Causal analyses in social media, for example, using propensity score matching and causal graphs.
  • Identifying natural experiments and using them to understand causal inferences
  • Identifying population, reporting and other biases in social media
  • Applications and domain-specific explorations
  • Novel methods for preserving privacy
  • Ethical codes and implications

Doctoral feedback session: If you are a PhD student doing research that uses causal inference methods on social media and user generated data, we are planning to incorporate a session for a doctoral feedback during the symposium. Please submit an abstract and let us know that you would like to be considered for this. Thanks to Nathan Matias for helping organize this effort!