PRTA:Joint extraction of medical nested entities and overlapping relation via parameter sharing progressive...

Area of Science:

• Natural Language Processing
• Medical Informatics
• Machine Learning

Background:

• Analyzing electronic medical records involves entity and relationship extraction, often treated separately.
• Existing models lack consistency due to isolated processing of nested entities and relationships.
• This separation hinders comprehensive information extraction from medical texts.

Purpose of the Study:

• To propose a joint medical entity-relation extraction model for improved consistency.
• To develop a model that simultaneously processes nested entities and relationships.
• To enhance the accuracy of information extraction from electronic medical records.

Main Methods:

• A joint model named Progressive Recognition and Targeted Assignment (PRTA) was developed.
• PRTA utilizes shared information from sequence and word embedding layers for simultaneous training.
• Novel strategies include a compound triangle for nested entity recognition and adaptive multi-space interaction for relationship extraction.

Main Results:

• The PRTA model demonstrated superior performance over state-of-the-art methods on multiple datasets.
• The method effectively addresses challenges posed by nested entities and overlapping relationships.
• Experiments were conducted on the Private Liver Disease Dataset (PLDD) and public datasets (NYT, ACE04, ACE05).

Conclusions:

• The proposed joint model significantly improves medical information extraction accuracy.
• PRTA offers a robust solution for handling complex entity and relation structures in medical records.
• This approach enhances the utility of electronic medical records for clinical analysis and research.