Mining impactful discoveries from the biomedical literature

Area of Science:

• Biomedical Informatics
• Computational Biology
• Data Science

Background:

• Literature-based discovery (LBD) uses text mining to find research connections.
• Current LBD evaluation methods are underdeveloped, hindering progress.
• A lack of robust benchmark datasets impedes LBD application development.

Purpose of the Study:

• To present a novel method for mining past discoveries from biomedical literature.
• To address the critical need for improved evaluation datasets in LBD.
• To facilitate the development of more advanced LBD techniques.

Main Methods:

• A new method leverages discovery impact and descriptive statistics to identify temporal relation surges.
• The approach analyzes the prevalence of concept relations over time.
• Method validity is assessed against the state-of-the-art "time-sliced" method.

Main Results:

• The method successfully collects a substantial volume of time-stamped discoveries.
• This collection serves as a valuable resource for LBD evaluation.
• The findings demonstrate a viable solution to the inadequate evaluation problem in LBD.

Conclusions:

• The developed method provides a large dataset of time-stamped discoveries for LBD evaluation.
• This resource can help overcome the long-standing issue of poor LBD evaluation.
• The approach may enable more sophisticated LBD methods and serve as a tool for systematic reviews.