Behind Malta’s Election Social Activity Scoreboard

05-06-2017

Following the success of ICON and Minely‘s collaboration to develop a Social Activity Scoreboard for Malta’s 2017 Election, Minely has given some insight on how, with the use of artificial intelligence, it analysed and segmented the social activity and data to produce the scoreboard.

AI and Machine Learning

AI is the concept of machines being able to carry out tasks in a way we would consider “smart”. It is an attempt to create a mechanical brain. Early computers were conceived as logical machines, designed to be capable of arithmetic and memory. As we advanced in technology our concept of what defines AI changed. Rather than solving complex calculations, AI is aiming towards replicating human decision making and differentiation.  This is the approach of machine learning, by learning for themselves as a human would.  Machine Learning applications can read through the text and determine whether it is conveying a positive or negative a message.  This is done by discovering patterns within the data. Such was the approach used in producing the Social Activity Scoreboard.

 

Approach and Process

Both of Malta’s largest political parties were analysed in terms of active users for the past few months. The key was to find patterns within the social data to extract the sentiment preference for both political parties. This process was coupled with a noise filter to exclude any stories which were not directly related to the political campaigns. From the charts, we can clearly see the development of sentiment for both parties, most notably as PN gained more users through the past months.

This was achieved using a four step process.

  • The first step was to build a user list of followers for both parties by extracting data on liked posts from various political Facebook pages.
  • The next step was to identify the unique users by assigning a score based on the number of likes made by each individual.
  • From there the focus turned to each party’s post across each media channel based on each unique user’s number of likes.
  • The final step is when AI came into play using a machine learning algorithm to partition the posts into clusters. The clusters, in this case, being PN and PL posts and non-political posts.

This project was tested for pin-point accuracy by manually analysing a sample dataset of posts. Any posts with a low volume of reactions were excluded due to a high impact on ratios. Alternative methods were considered such as analysing the text for sentiment but since the Maltese NLP (natural language processing) tools are limited, this option was ruled out.

More in-depth reading can be found on Minely’s article regarding the Scoreboard and the process.