Ciencia España , Salamanca, Jueves, 23 de noviembre de 2017 a las 13:29

A tool to classify political opinions on Twitter

The University of Salamanca has developed a real-time tuit classifier that works in the cloud through an automatic learning process

FGUSAL/DICYT Social networks have become a prime space for public debate and among them Twitter especially stands out in the field of politics. Some computer tools already analyze trends and opinions. The University of Salamanca has developed a messages classifier that allows to know in real time the positive and negative evaluations of the users.


It is a real time political tweets classifier that allows distinguishing between positive, neutral and negative feelings. We have developed it from the supervised feelings analysis technique, which uses machine learning algorithms, based AI",explains to DiCYT Carlos Arcila Calderón, researcher at the Department of Sociology and Communication.


In an initial phase, researchers adapted from English to Spanish an existing Python code.
To do this, they trained the system through a corpus that collected examples of political tweets in Spanish which have been manually classified.

Now they have uploaded it to the cloud with the Apache Spark open source platform, a development that was not free access in any language. This will allow them to analyze efficiently and safely a large amount of data from anywhere.

The new AutoCop prototype will address itself massive data analysis processes, also known as Big Data. As it is hosted in the cloud, it overcomes personal computers limitations as it is able to process, manage and analyze data coming from Twitter, very fast, in real time and in a distributed way.


"Not only it´s faster, but it is more stable because it does not depend on the difficulties we may have at home or in the office, such as the power going out or the computer crashing, in the cloud we can distribute and execute the analysis with a lot of stability and speed, "says Arcila.


To develop this second phase of the project, researchers of the Observatory of Audiovisual Content from the University of Salamanca have taken part in the Proof of Concept call of the General Foundation of the University of Salamanca, within the TCUE program of the Junta de Castilla y León co-financed with FEDER funds.


Real time graphics

"One of the main advantages of AutoCop, both in its local and distributed version, is that we can visualize "Tweeter real time feelings", that is, we have graphics that indicate what feeling exists in a certain moment towards a certain political party or towards a certain idea ", highlights Arcila.


Given the characteristics of this machine learning classifier, the creators of AutoCop consider that it can be applied both for scientific research and for being transfer to other sectors such as journalism, political consulting and the parties themselves.


"Our software and know-how may be useful in different sectors, for political communication, corporate social responsibility, including marketing and public health", Félix Ortega Mohedano, one of the project's researchers. "Twitter may be considered the arena for nowadays public dialogue. So that, AutoCop allows us to study what´s happening in social media through scientific methodologies and Big or 'medium-sized data', he adds.


Through this project, knowledge in subjects such as politics, communication and sociology go hand in hand with the most advanced technologies and artificial intelligence to better understand reality.