ParaBTM: A Parallel Processing Framework for Biomedical Text Mining on Supercomputers.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Data Science

Background:

• Biomedical literature analysis relies heavily on text mining of unstructured data.
• The exponential growth of biomedical literature presents significant big data challenges for processing efficiency.
• Existing text mining methods struggle to keep pace with the volume of scientific publications.

Purpose of the Study:

• To address the big data challenge in biomedical text mining by leveraging parallel processing.
• To develop and evaluate a novel parallel text mining framework, paraBTM, for supercomputer environments.
• To enhance the efficiency and scalability of extracting information from large biomedical corpora.

Main Methods:

• Developed paraBTM, a framework enabling parallel text mining on the Tianhe-2 supercomputer.
• Implemented a low-cost, effective load balancing strategy to optimize parallel processing.
• Evaluated paraBTM performance using named entity recognition (NER) tasks on diverse datasets.

Main Results:

• Parallel processing with paraBTM significantly improved processing efficiency in most evaluated scenarios.
• The proposed load balancing strategy demonstrated simplicity and effectiveness in maximizing parallel task execution.
• paraBTM achieved substantial performance gains, showcasing the benefits of supercomputing for text mining.

Conclusions:

• Parallel processing on supercomputers, facilitated by frameworks like paraBTM, is a viable solution for big data challenges in biomedical text mining.
• The paraBTM framework and its load balancing strategy offer a scalable and efficient approach for biomedical information extraction.
• The framework's applicability extends beyond named entity recognition to various other biomedical text mining tasks.