Zuhlke has been collaborating with peacebuilding charity International Alert since the summer of 2014 in a pro-bono project to look at opportunities for gaining insights and fresh perspectives on the situations in which they operate using social media and other forms of digital data.
Alert is one of the world’s leading peacebuilding organisations, with nearly 30 years of experience promoting peace. Its primary goal is to help reduce conflicts within nations (as opposed to those between nations), and it is currently active in over 25 countries around the world. Much of their work involves staff on the ground collaborating with local organisations and populations to understand the causes of conflict and to initiate and support programmes addressing these causes.
We have focussed initially on Nigeria, one of the largest countries in which the charity operates, and one with a high rate of English usage. We have built a small Azure cluster running the Hortonworks Data Platform collecting and analysing locally produced news articles and blog posts as well as tweets originating in Nigeria or related by keywords. A web-site in the cluster allows Alert staff to configure and run a variety of reports on the data including sentiment analysis, heat maps of tweets, and cluster graphs showing connections between actors mentioned in articles.
Some interesting results were obtained during the recent delayed Nigerian presidential elections, when the incumbent Goodluck Jonathan was unexpectedly defeated by the opposition leader Mohammed Buhari. We ran a lexical sentiment analysis algorithm based on recent research papers on sets of tweets referencing the two main candidates, which indicated that former president Jonathan would have had a much better chance of winning if he had held the elections at the originally planned date around the middle of February. The ratio of positive to negative sentiment was highest for him in the run up to the election, then interest dropped rapidly after the postponement. Whilst still broadly positive, the levels didn’t recover for the election finally held at the end of March.
In contrast, the sentiment for Buhari was biased toward the negative leading up the February election date, but changed markedly as the new date approached. A small spike around the 26th Feb correlates with a successful press conference he held in London. However by March 24th, positive sentiment strongly outweighs negative sentiment and this matches the eventual election result.
Another interesting sentiment report was run on corruption and bribery. Whilst the sentiment was consistently negative, there is an interesting spike in opinion at the point when the elections are postponed.
Simple analyses such as these above seem to exhibit a significant correlation with real world events. Next steps for the project include expanding the range of analyses available, and looking to see if it is possible to identify trends around topics of particular interest to peacebuilding activities, and even to see if it possible to discern the local impacts that peacebuilding agencies may be having. The signal to noise ratio makes this a real challenge for Zuhlke and Alert. Stay tuned for further reports from the frontline!