Description
Language models like BERT (an others based on transfer learning) have created an enormous interest in natural language processing (NLP) during the last years. As a recent application, we want to introduce sentence embeddings in this talk. They can be implemented easily and are very useful for a variety of applications. Details can be found on sbert.net.
In the talk, we will focus on the theory shortly and then turn to applications with examples for re-ranking, cross-encoders and comparing meaning of sentences across different languages.
Combined with other methods, sentence embeddings can be used to automatically classify unlabeled data or extend partially categorized data. We will explain both use cases. We encounter both scenarios frequently in our own projects and so far they could only be solved with considerable manual effort.