Materials

The tutorial follows a highly practical approach. The teaching materials fundamentally consist of Jupyter Notebooks on Google Colaboratory, which can be run totally on the cloud and do not require installation. We will also use slides as support material.

The notebooks corresponding to the lessons are available directly in Colaboratory:

Note that some of the lessons use considerably large datasets and models. It is advisable to load them from the notebooks in advanced. All the notebooks include the necessary code to do so.

You can also access directly to each lesson by clicking in the icons below:

Lesson 1: Capturing meaning from text as word embeddings.
Lesson 1a: Fine-tuning pre-trained language models for text classification
Lesson 2: Knowledge graph embeddings.
Lesson 3: Building a vecsigrafo – generating hybrid knowledge representations from text corpora and knowledge graphs.
Lesson 3a: From vecsigrafo to transigrafo – Using neural language models based on transformers to build hybrid representations
Lesson 4: Evaluating vecsigrafos beyond visual inspection and intrinsic methods
Lesson 4a: Assessing relational knowledge captured by embeddings
Lesson 5: Vecsigrafos for curating and interlinking knowledge graphs.
Lesson 6: Exercise – Correlating language in classic literature.
Lesson 7: Application – Fake news and deceptive language detection.
Lesson 7a: Semantic claim search
Lesson 7b: Application – (Dis)credibility propagation
Lesson 8: Application – Scientific information management.
Lesson 9:  Application - Classifying scientific literature using surface forms
Lesson 9: Application – Classifying scientific literature using surface forms