MeSHProbeNet: a self-attentive probe net for MeSH indexing.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Information Retrieval

Background:

• MEDLINE is a crucial bibliographic database for biomedical research, indexed using Medical Subject Headings (MeSH) for knowledge discovery.
• Manual MeSH indexing is time-consuming and costly, driving the need for automated solutions.
• Existing automated MeSH indexing systems often face efficiency challenges due to reliance on intermediate model outputs.

Purpose of the Study:

• To develop an efficient and accurate end-to-end framework for automatic MeSH indexing.
• To introduce MeSHProbeNet, a novel system utilizing deep learning and self-attentive MeSH probes.
• To improve the speed and reduce the resource requirements of MeSH term assignment.

Main Methods:

• Developed MeSHProbeNet, an end-to-end deep learning framework employing self-attentive MeSH probes.
• Each probe is designed to extract specific aspects of biomedical knowledge from articles.
• A unified classifier recommends MeSH terms based on extracted information, ensuring interpretability at the word level.

Main Results:

• MeSHProbeNet achieved first place in Task A of the 2018 BioASQ challenge.
• The model demonstrated superior performance across all F-measure metrics (Example Based, Macro, Micro, Hierarchical, LCA) compared to state-of-the-art models like MTI and DeepMeSH.
• The system effectively extracts comprehensive biomedical knowledge, offering word-level interpretability.

Conclusions:

• MeSHProbeNet offers a time and space-efficient solution for automatic MeSH indexing.
• The self-attentive probe mechanism allows for detailed extraction of biomedical concepts.
• This framework significantly advances automated biomedical literature indexing and knowledge discovery.