I could finish the prototype implementation that allows to search for a hashtag and filter Twitter messages for sentiment, credibility and media type. It was quite a big task as i had to implement a proper structure to allow a modular design and implement a server and client from scratch (with libraries of course, e.g react.js and node.js)

I could implement a SVMClassifier that works with node-svm and got trained with three models from my experiments with the MediaEval data, based on features from the user study. The best approach has an F1-Score of 0,68 and is within the result range of approaches that employ the same MediaEval data with textual features. To compare it with a real life example: A random guess would allow you to have a rate of 0,5 (50/50) to correctly guess a Tweet as credible or not.

Not an extreme enhancement, but an enhancement. However, the goal was to create a modular system design that allows to implement further scientific knowledge!