É«É«À²

For his Master's thesis in Information Systems Science, Sippo Rossi developed a machine learning model for detecting Twitter bots. With this model, he investigated whether fake accounts were used to influence voters during the 2019 Finnish parliamentary elections.

A bot is a computer program that performs certain tasks according to defined instructions, and at least partially independently, i.e. without human intervention.

‘The model I have developed can effectively identify simple bots based on Twitter profile metadata. According to my model, more than 30 percent of some of Finland’s top politicians’ followers appear to be bots. This figure is considerably higher than Twitter's own estimates of the global share of bots,’ says Sippo Rossi, who completed his Master's thesis at Aalto University School of Business.

The study analysed data on 550,000 Twitter accounts using network analysis. The data used in the study was extracted from Twitter during March 2019. The data consists of Twitter followers of 14 well-known Finnish politicians. The biggest concentrations of bots were found among the followers of politicians such as Alexander Stubb and Pekka Haavisto.

The origins and goals of the bots are unclear

According to Rossi, in most cases the origin and purpose of the bots remained unclear. It is therefore difficult to determine what proportion of them has been deliberately created to increase the number of politicians' followers.

‘Seemingly, many bots follow Finnish accounts purely on Twitter's recommendations, which leads to a cycle in which the most followed politicians collect extensive botnets.’

According to the Master's thesis, the impact of bots on Finland's political environment on Twitter is limited despite the fact that bots clearly enhance the visibility of certain politicians. These results are in line with the findings of a recent report by the Finnish Security Intelligence Service, according to which foreign countries did not attempt to influence the elections. The study shows that Twitter still has great difficulty in removing even simple and easily detectable bots.

Link to the Master’s thesis: