This approach consists on a practical semantic similarity experiment in order to automatically grade students answers, the data could be found in https://www.kaggle.com/smiles28/semeval-2013-2-and-3-way , the only thing you should change over script are the paths in preprocess steps, then undestand the code in order to start fine tuning on your own parameters atack, it also offers some preprocess stages that we din't apply for time issues, but perfectly applicable, some experiment results are shown on the repository to address if the approach is suitable for your purposes.
To understand the overall system we add the presentation file were we explain the overview, in general words we aim to provide understandable Deep Learning models in order to provide suitable technology for automatic answer scoring systems. Those systems are highly demanded on learning areas. We recorded the presentation and you can ear it here https://drive.google.com/open?id=18dIn72VsvGVBD27hWjsfPKsP8G0BmdfQ.
Note: the model is dependent on external libraries that will be installed, so the first run may take a while on installing the dependencies.
Edgar Andrés and Mohamed Yassin.