The vast majority of information being made available online is contained in unstructured text (news/blog articles, scientific publications, etc.). The goal of information extraction is to develop methods to bring this information into a structured form (knowledge bases). Due to the complexity of this challenge, it is typically broken down into multiple steps, such as named entity recognition (NER), entity linking, relation extraction, event extraction, etc.
The goal of relation extraction is to detect relations between entities (such as a person or organization) mentioned in a text. Typically, this task is evaluated on text passages in which entities have been labeled manually. In joint entity and relation extraction this is not the case. In-stead, both the entities as well as the relations between them have to be identified in an input text. This removes the necessity of labeling entities by hand and while this makes the task more difficult initially, multi-task setups can often improve performance on end-to-end tasks.
The focus of the proposed thesis is to compare existing approaches (such as ,) for joint entity and relation extraction and develop, implement, and test own approaches.
Hands-on experience in machine learning, no fear to implement neural network models (under guidance of the supervisors).
Ausschreibung: Download (pdf)