calculating the similarity of views in Sonar


As an experiment, we decided to see what similarity between users looks like and whether information about this similarity is interesting to users.

In Sonar, users are given the opportunity to answer short (one-question) surveys to test the similarity of their views with the positions of different political parties.


We start with a multidimensional space. Each survey or poll is one dimension. However, such a solution is not very practical (due to laboriousness and difficulty in imagining it). Therefore we decided to use PCA (Principal Component Analysis), which is a classical algorithm from dimension reduction family. It transforms our data in such a way that only five dimensions remain, with the least possible loss of information in relation to their original form.


Our contribution to Data Driven Journalism: we were the first in the world to create an algorithm to visualize the similarity between users of a news service based on polls and surveys. We transformed the extracted data to embed it in five dimensions, with as little loss of information from its original form as possible.

Interested in an experiment to extract key data and dependencies in your product?

Required *