Discovering Outcomes via Propensity Score Analysis of Social Media Timelines


Event details

Date 10.12.2015
Hour 11:0012:00
Speaker Emre Kiciman, Microsoft Research Redmond
Category Conferences - Seminars
People make decisions every day; some are significant and life-changing decisions and others minor and mundane. Relying on their personal experience and knowledge, individuals rightly or wrongly choose what to do based on expectations of their decision’s outcomes.  What if we could tell everyone what happened when others were in the same situation?  With 10-100s of millions of people regularly reporting the details of their real-world experiences in social media, we believe there is an opportunity to mine the likely outcomes of situations and actions to help people make more informed decisions.

We present a framework for extracting such outcomes from personal experiences reported in social media timelines.  Applying stratified propensity score analysis, we isolate the outcomes of specific events from observed confounding factors.  We benchmark the quality and coverage of results extracted from 3 months of Twitter data for 39 situations in a wide-range of domains including social, health and business topics.  We also demonstrate the application of this analysis technique in deeper case studies, including a study of the precursors of shifts towards suicide ideation among pseudonymous posters in Reddit support communities.

Practical information

  • General public
  • Free


social media causal inferences action outcomes

Event broadcasted in