Mapping Twitter Chatter in Iraq

The rapid expansion of the Islamic State (IS) across different parts of Iraq and the dramatic escalation of violence in late 2013 has caused millions of Iraqis to flee their homes. Here we attempt to understand the relationship between chatter and buzz on Twitter and the IS movement through Iraq. We use hundreds of millions of tweets related to IS to investigate chatter volume, sentiment of tweets involving IS related hashtags in both English and Arabic.

Because we are interested in movement as individuals flee IS, we collected tweets containing #ISIS and داعش‎# . The figures below describe this data and how it relates to movement.

Detailed Views on Mosul Offensive and Fallujah Offensive

Scaled Dynamics of Chatter

In order to gain insights in how IS affects movement, we created localized signals from the #ISIS and داعش‎# tweets. This map shows the number (volume) and sentiment of ISIS tweets mentioning a location within a given governorate over time.

This map also shows events that occurred within each governorate. Click on each bubble to learn more about what was happening in a governorate in a given month.


Chatter, Sentiment, and Displacement

The bar charts below allow you to view the raw data over time. These data include all tweets, not just those mentioning a location. One interesting insight is that positive and negative sentiment are correlated, but while English and Arabic tweets are also correlated, there each follow slightly different temporal patterns. . This highlights how important it is to capture diverse signals. Another insight is that English is always more negative in #ISIS tweets, while Arabic tweets show more variation, leaning toward more positive. It should be noted, however, that sentiment, or emotional overtones, is not the same as stance. Finally, there are spikes in the number of tweets after major attacks or violent events.



** Select three buttons to change data on Bar Graph.


** Slide on either end to adjust timescale.

** Select three buttons to change data on Bar Graph.


** Slide on either end to adjust timescale.





While signals on the bar chart are difficult to compare because of differences in scale, this line chart allows for comparison by displaying the percent change each month from the previous month. One interesting takeaway is that the number of new families displaced from month to month does not change as drastically as social media signals. The comparison between English and Arabic signals is more clear in this chart. There is a clear correlation, but also considerable variability between the two. This chart also highlights that while positive and negative sentiment in English are closely tied, there is divergence between the positive and negative signals in Arabic. This suggests that in Arabic, each may be capturing something slightly different.

These charts provide a few take away messages:

  • English sentiment is always more negative than positive.
  • Arabic sentiment changes through time in terms of whether ir is more positive or more negative.
  • Different spikes in volume map directly to different large IS attacks (Ramadi & Mosul) or events.
  • Arabic volume and sentiment seem to be leading movement indicators of where people are going. English volume is also a leading indicator, but not as clear.
  • Strong relationships exist between tweet volume in English and Arabic, tweet sentiment in Arabic, death counts, and movement. This means that these variables are reasonable indirect indicators.