Nested Named Entity Recognition via Language Model Based Neural Architecture
Expert: Sefer Baday
Most of the models on NER focus on flat entities that are able to associate a text span to only one class, which decreases the generalizability of the models and restricts the extraction of comprehensive information. Nested entities are a part of the nature of the language and using nested entities, helps better abstraction of information. Usage of nested entities would reveal the fact that locations tend to be used for naming organizations. While this kind of situation will only increase the level of complexity for the NER model for flat entity recognition.
The company will perform their models on a novel Nested-NER dataset called LitBank. All the sources used in the Litbank dataset come from public domain texts in Project Gutenberg. All the texts were originally published before 1923. Besides, the Artiwise team is building a new dataset from Turkish news articles (especially in the Economy domain) under a granted TUBITAK TEYDEB 1501 project. The company has a plan to get our dataset ready during this project time and apply the model to this data set.
Please click here to view the results of the case study carried out with the company.
Please click here to reach the success story.