RAND Corporation researchers developed an approach for analyzing social media data to derive insights about chemical incidents and conducted a proof of concept of that approach by applying it to the case of chemical weapons use in Syria (2017–2018). The procedure, a four-step process, showed promise. They recommend that the Defense Threat Reduction Agency initiate three activities to further the development of the procedure.
Using Social Media to Extract Information About Chemical Weapons Incidents
A Methodology and Demonstration of Concept from the Civil War in Syria
- When a chemical event happens, what information about it tends to appear in social media, and what forms does that information typically take?
- How can a computational approach enable rapid detection of chemical weapons incidents buried among millions of social media posts?
- How can a blended computer-human approach improve on a fully automated approach?
- What do leaders need to know to implement a social media analysis capability?
Policymakers across the federal government have begun to recognize the potential of social media as a source of information and have commissioned studies to explore how social media can improve disaster situational awareness, influence public opinion, augment traditional data sources, and counter disinformation. In this project, RAND Corporation researchers developed an approach for analyzing social media data to derive insights about chemical incidents and conducted a proof of concept of that approach by applying it to the case of chemical weapons employment in Syria between 2017 and 2018.
They identified a four-step process: (1) Identify operationally relevant factors and examine known events to find incident indicators, (2) develop a feed of social media data, (3) conduct automated daily scans for elevated keyword use in Twitter data, and (4) analyze posts to verify detection and extract information.
The procedure showed promise. Based on the analysis, it is recommended that the Defense Threat Reduction Agency initiate three activities to further the development of this procedure.
Twitter, Facebook, and YouTube post text may include chemical attack descriptor keywords, place names, and broad characterizations of the agent used
- It may also include reactive language, such as expressions of anger, lamentation, or religious invocation.
- Images, videos, and links may provide more incident detail, as well as ways to verify authenticity.
Scanning a Twitter sample for chemical incident–descriptor keyword and reactive-language surges can provide analysts with a low-latency alert that an event may have happened
- This can direct them toward posts they can use to adjudicate whether it has happened.
- The researchers scanned Twitter during our research for this report, but choosing which platform to scan is a complex decision with many considerations.
Human intelligence may be better able to cope with the irregularity of social media data and better poised to use sophisticated inference and verification techniques to investigate incidents
- If computation can filter data volume down to something manageable, the human intelligence component can generate deeper insights from it.
The information value of social media methods varies from place to place, depending on characteristics of the population, relevant state actors, social media platforms, and analyst capabilities
- Implementing this method also requires supporting proportional staffing levels and taking steps to protect staff from posttraumatic stress.
- Develop resource pooling agreements with other federal agencies.
- Build an extensive chemical weapons keyword list and rigorously test the list's detection effectiveness against best available intelligence.
- Conduct an exercise with federal computers and staff, scanning for easily verified types of events.
Table of Contents
Social Media Analysis of Chemical Weapons Incidents in Syria
Conclusions and Policy Implications
Additional Technical Tables
Quality of the Social Media Environment
Estimating Analyst Level of Effort Requirements
Protecting Analysts from Secondary Trauma
Social Media Event-Detection Literature Review