Distinguished representation of identical mentions in bio-entity coreference resolution

Area of Science:

• Biomedical text mining
• Natural Language Processing
• Bioinformatics

Background:

• Bio-entity Coreference Resolution (CR) is crucial for biomedical text mining.
• Identical mentions with similar representations pose a significant challenge in CR.
• Existing neural network models may introduce noise, hindering the distinction of identical mentions.

Purpose of the Study:

• To develop a model that effectively distinguishes similar or identical text units for improved coreference resolution.
• To address the challenge of differential representation in Bio-entity CR.
• To reduce noise and enhance semantic information capture in biomedical text mining.

Main Methods:

• Proposed a context-aware feature attention model.
• The model adaptively exploits features based on different contexts.
• Enabled reduction of text noise and effective capture of semantic information.

Main Results:

• Significant improvements in coreference resolution and mention detection on BioNLP and CRAFT-CR datasets.
• Demonstrated superior performance in the differential representation and coreferential linking of identical mentions.
• Empirical studies confirmed the model's effectiveness.

Conclusions:

• Identical mentions present difficulties for current Bio-entity CR methods.
• The proposed context-aware feature attention model offers superior performance.
• This model enhances both coreference resolution and mention detection, benefiting downstream tasks.