With the progressive increase of technology in the daily life, the use of the social networks has become into a relevant tool for politicians in the aim of catching new possible voters. However in the side of the citizen, the high amount of information can produce a lose of perspective.
This project is focused on the design and development of a system able to classify
political messages according to their bias into partisan or neutral. For the development of the system it is used the dataset Political-Social-Media imported from Kaggle. Thus, the system is trained with Facebook and Twitter messages, and will be used to predict new samples of these social networks.
For the design of the system were used two different approaches: The Bag of Words model and a basic Neural Network system. In addition to this, it is also developed a third solution based on the combination of the two previous.
Finally the system is tested with a different set of samples, used only to verify the
behaviour of the system. Once are selected the best models for Twitter and Facebook, the final result is implemented in a plugin. For this task, it is employed Senpy that will allows to classify new samples in real time, as well to share the system in an easier way due to the use of linked data.
The last step of our project is to extract the main conclusions of the work and join them with the new technologies learned, in order to stipulate possible future lines of work.