Extending TextAE for annotation of non-contiguous entities

Area of Science:

• Bioinformatics
• Natural Language Processing
• Computational Biology

Background:

• Named entity recognition (NER) is crucial for biomedical information retrieval.
• Current NER tools often fail to recognize non-contiguous entities, common in biomedical lists.
• This limitation hinders accurate text annotation and dataset creation for machine learning.

Purpose of the Study:

• To extend the TextAE platform for visualizing and annotating non-contiguous biomedical entities.
• To enable users to add subspans to existing entities and edit annotations involving non-contiguous entities.
• To address the limitations of existing text annotation systems in handling complex entity structures.

Main Methods:

• Extended the TextAE platform to support non-contiguous entity visualization and annotation.
• Integrated new functionality for adding subspans to existing entities.
• Enabled editing of relation annotations involving non-contiguous entities and supported PubAnnotation format import/export.

Main Results:

• Successfully enabled visualization and annotation of non-contiguous entities within the TextAE platform.
• Facilitated easier editing of entity and relation annotations, including those with non-contiguous spans.
• Quantified the prevalence of non-contiguous entities in biomedical literature, highlighting missed information by current systems.

Conclusions:

• The enhanced TextAE platform effectively addresses the challenge of annotating non-contiguous biomedical entities.
• This advancement improves the quality of biomedical datasets for machine learning and information extraction.
• The findings underscore the significance of supporting non-contiguous entities for comprehensive biomedical text mining.