ASP-HR: An Adaptive Spatial Perception and Hierarchical Reasoning mechanism for document-level biomedical relation...

Area of Science:

• Biomedical informatics
• Natural Language Processing
• Machine Learning

Background:

• Document-level biomedical relation extraction is vital for advancing disease diagnosis and drug discovery.
• Current methods struggle with static entity representations, document structure, and lack of global consistency.
• These limitations hinder the accurate mining of complex relationships from biomedical literature.

Purpose of the Study:

• To introduce a novel model, Adaptive Spatial Perception and Hierarchical Reasoning (ASP-HR), for improved document-level biomedical relation extraction.
• To address the limitations of existing methods by incorporating dynamic entity weighting and hierarchical reasoning.
• To enhance the accuracy and robustness of extracting biomedical relationships from text.

Main Methods:

• Developed a three-level progressive architecture featuring the Dynamic Distance-Aware Entity Interaction Enhancement (DDA-EIE) module.
• Implemented the Multi-Relation Graph Attention Reasoning Network (MR-GARN) for deep semantic dependency mining.
• Integrated the Dual-Stage Inference Enhancement (DSIE) module for robust prediction through bias correction.

Main Results:

• ASP-HR achieved superior performance on the CDR dataset with an Overall-F1 of 88.9%, outperforming the state-of-the-art by 1.8%.
• The model demonstrated strong results on the GDA dataset (Overall-F1: 84.9%) and BioRED dataset (Overall-F1: 61.8%).
• ASP-HR significantly improved Intra-F1 and Inter-F1 scores, indicating enhanced ability to capture both within-sentence and cross-sentence relationships.

Conclusions:

• The proposed ASP-HR model effectively addresses key challenges in document-level biomedical relation extraction.
• Adaptive spatial perception and hierarchical reasoning mechanisms are crucial for enhancing model performance and robustness.
• ASP-HR represents a significant advancement in the field, offering improved accuracy for biomedical knowledge discovery.