SoBigData Articles

Understanding peacefulness through the world news

Could data science help measure peacefulness and understand the factors that influence it? Could we anticipate the level of peacefulness before official sources publish their estimations?

Recently, we have presented a study [1, 2] that demonstrates how using Global Data on Events, Location, and Tone (GDELT) and applying machine learning techniques we can predict the Global Peace Index score (GPI) [5]. GPI captures the peacefulness of the world at a yearly level. Therefore, we show how news data from GDELT help capture GPI at a monthly frequency, anticipating the yearly index value.

To expand our understanding of the factors that influence the present and future peacefulness we use the SHapley Additive exPlanation (SHAP) methodology [3, 4]. In particular, we apply SHAP methodology to provide model transparency on the factors that contribute to higher or lower levels of peacefulness.

SHAP shows the most important factors that influence GPI at a global level and at a local level. For example, in the United States, the most important factors that contribute to the estimation of GPI, as calculated for the training period between April 2014 and March 2020, indicate a country's profile of a strong player in the military, socio-economic, and political foreground (see Global variable importance plot on Figure 1).  


Fig. 1: Global variable importance plot for the United States XGBoost model. The barplot orders the variables based on their importance in the estimation of the GPI score. Overall, we show that the variables indicate a country profile of a strong player in the military, socio-economic, and political foreground.
 
In addition, we present the prediction of the GPI value in June 2020. We choose this month, since the murder of George Floyd, took place on May 25, 2020, in the US. Other events followed this extreme event for the whole of June 2020, such as protests, and it provoked an amount of news concentrated on the topic. Thus, it would be interesting to study this month’s level of peacefulness and understand the factors that drive the GPI estimation. Fig. 2 displays the local plot for the GPI prediction in June 2020. We observe that the estimated GPI is 2.30 which corresponds to the model output value for 3-months-ahead prediction. This value indicates that GPI will remain stably high in June 2020 compared with the last official yearly (2.31), and the median GPI value of the previous three years (2.34). Additionally, the red and blue arrows correspond to the factors which increase and decrease the GPI, respectively. For example, news of the category “Protest violently, riot” increases GPI. Indeed, in June 2020, the news was concentrated on a series of protests, followed by the murder of George Floyd against police brutality and racism. The model reveals that protesting in the United States contributes to the improvement of various socio-political situations, and as a consequence to peace-building.
Fig. 2 Local plot for the United States XGBoost model. It presents the estimation of the GPI for June 2020 and the most important variables that the model uses for the estimation. The red and blue arrows are the variables that increase or decrease the GPI respectively.
 

Considering that the global society is under socio-economic and political crisis and instability, policy-makers and peace-builders need to have frequent updates of peacefulness, and of the factors which influence its levels. Additional understanding of peacefulness could empower them to timely react to applying adequate policies, preventing detrimental societal effects, and contributing effectively to social well-being and progress.

 

[1] V. Voukelatou, L. Pappalardo, I. Miliou, L. Gabrielli, F. Giannotti, “Estimating countries’ peace index through the lens of the world news as monitored by GDELT” Proceedings of IEEE International Conference on Data Science and Advanced Analytics (DSAA2020)

[2] V. Voukelatou, L. Pappalardo “How GDELT is changing how we measure peacefulness?” http://sobigdata.eu/blog/how-gdelt-changing-how-we-measure-peacefulness

[3] Lundberg, Scott M., Gabriel G. Erion, and Su-In Lee. "Consistent individualized feature attribution for tree ensembles." arXiv preprint arXiv:1802.03888 (2018).

[4] Lundberg, Scott, and Su-In Lee. "A unified approach to interpreting model predictions." arXiv preprint arXiv:1705.07874 (2017).

[5] The Institute for Economics and Peace, “VISION of HUMANITY.” http://visionofhumanity.org//, 2017.

Written by: Vasiliki Voukelatou

Revised by: Luca Pappalardo

SoBigData++ Exploratory: Demography, Finance, and Economy 2.0

Publication date

Jul 28, 2022