JCBIE

JCBIE performs joint continual learning to extract biomedical entities and relations from corpora for construction and expansion of large-scale Biomedical Knowledge Graphs (BKGs).

Key Features:

• Joint Continual Learning Approach: Integrates joint learning and continual learning strategies to incrementally learn new entity types and relations while retaining previously acquired knowledge for BKG construction.
• Dual Encoder Architecture: Employs two separate encoders rather than a single hard-parameter sharing encoder to mitigate feature confusion and provide distinct feature representations for entities and relations.
• Entity-Augmented Inputs: Uses entity-augmented inputs to enable interaction between named entity recognition (NER) and relation extraction tasks, improving entity identification and relation linking.
• Cross-Corpus Generalization Evaluation: Implements an evaluation mechanism that measures cross-corpus generalization errors to assess robustness across different corpora.

Scientific Applications:

• Biomedical Knowledge Graph construction: Facilitates extraction of diverse biomedical entities and relations to build and expand large-scale BKGs.
• Downstream biomedical research: Supports applications such as drug discovery, disease modeling, and personalized medicine by providing structured entity and relation data.

Methodology:

Empirically compares joint learning and continual learning strategies; implements a dual encoder architecture; employs entity-augmented inputs to enable NER–relation extraction interaction; and uses a cross-corpus generalization evaluation to measure generalization errors.