This project is undertaken in the framework of the lecture and seminar Knowledge Discovery at the Karlsruhe Institute of Technology (KIT). Its main research question is to examine whether a combination of text and image classification has the potential to yield better results than any of the two classifiers individually. Hence, our task involves first of all finding an interesting use case and generating the necessary data set. Secondly, we train two stand-alone classification models as to investigate in a third step how to combine these most suitably. The objective is to obtain improved classification performance with such a bi-modal classifier.
Our project unfolds around the field of computational sentiment analysis in general and around
extracting opinions expressed in social media towards the 2016 California Primaries in particular.
Primaries are pre-elections which are held in most US states as part of the parties’ elections of
their presidential candidates. At the time of undertaking this project, Donald Trump was already
certain to be the republican nominee while the democratic nominee was not decided upon yet. Thus,
our analysis focuses on determining the attitudes of social media users expressed during the rally
between Hillary Clinton and Bernie Sanders.
