Related Experiment Video
Updated: Sep 6, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
4.1K
Multi-probe attention neural network for COVID-19 semantic indexing
Jinghang Gu1, Rong Xiang2, Xing Wang3
1Department of Chinese and Bilingual Studies, The Hong Kong Polytechnic University, Hong Kong, China.
BMC Bioinformatics
|June 29, 2022
Summary
This study introduces a new AI framework, the multi-probe attention neural network (MPANN), for automatically indexing COVID-19 research. MPANN efficiently analyzes biomedical articles to predict semantic topics, aiding in managing the pandemic
Area of Science:
- Biomedical Informatics
- Artificial Intelligence in Medicine
- Information Retrieval
Background:
- The COVID-19 pandemic has led to a surge in scientific publications, necessitating efficient methods for literature curation and indexing.
- Traditional manual indexing using Medical Subject Headings (MeSH) is time-consuming and resource-intensive.
- Automatic semantic indexing is crucial for managing the rapidly growing body of COVID-19 research.
Purpose of the Study:
- To address the challenge of semantic indexing for the vast amount of COVID-19 literature.
- To develop and evaluate an automated framework for predicting semantic topics in biomedical articles related to COVID-19.
Main Methods:
- Construction of a novel COVID-19 Semantic Indexing dataset comprising over 80,000 biomedical articles.
- Proposal of a multi-probe attention neural network (MPANN) framework for semantic indexing.
- Utilizing a k-nearest neighbour based MeSH masking approach to generate candidate topic terms and incorporating them as probes into an attention-based neural network.
Main Results:
- The MPANN framework effectively extracts semantic features at both term and document levels.
- A linear multi-view classifier is employed for final topic prediction within the MPANN framework.
- The proposed method demonstrates proficiency in handling the semantic indexing problem for COVID-19 literature.
Conclusions:
- The MPANN framework shows promise in representing semantic features of biomedical texts.
- MPANN is effective in predicting semantic topics for COVID-19 related biomedical articles.
- The developed approach offers an efficient solution for the automatic indexing of pandemic-related scientific literature.
