Förderjahr 2019 / Stipendien Call #14 / ProjektID: 4563 / Projekt: Automated Identification of Fake Information in Social Media from Multimodal Data
So roughly a year is over again since I started to prepare for this final part of my thesis.
The last few weeks have been very interesting but also exhausting. I needed to overthink and rebuild a lot of my networks due to not recognisable problems. The main breaking point was the fusion part.
To get into more details: The main problem lies in the connection of each model ( Model 1, 2 ...). I need to get the output before the classification layer and take these information to feed it to the "fusion head". In detail, the problem was to load the pretrained weights after fine-tuning back into the ensemble model in order to beneftit from the learned information.
After overthinking and a lot of try and error, I finally managed it to make it work again.
So after fixing this issue the systematic experiments can start to evaluate my network.
Since my last blog post a lot of things happend in the field of fake news. Along with the spread of covid-19 a lot of fake news / fake information has been spread over the internet mentioning that covid is a fake, and/or these helps against it or not. These fakes are extreme harmful to the readers because it is very hard to verify these information. But also my misinformation detector would have problems with verifying it. Maybe this would be a topic for a project? ;)
The next two month, till the end of August are mainly focused on finishing my thesis and hand it in.
Stay tuned ;-)