A novel NIH research grant recommender using BERT.

Area of Science:

• * Computational Biology
• * Bibliometrics
• * Information Science

Background:

• * Securing research grants is crucial for academic career progression.
• * Identifying suitable funding opportunities, particularly from agencies like the National Institute of Health (NIH), is a significant challenge for researchers.
• * Existing methods for grant discovery are often time-consuming and inefficient.

Purpose of the Study:

• * To develop an automated recommendation system for National Institute of Health (NIH) grants.
• * To utilize researchers' publication data as a basis for personalized grant recommendations.
• * To leverage advanced deep learning techniques for improved accuracy in grant matching.

Main Methods:

• * Formulated grant recommendation as a classification problem.
• * Employed Bidirectional Encoder Representations from Transformers (BERT) to model complex relationships between publications and grants.
• * Conducted internal evaluations using grant citations and external evaluations involving researcher feedback and Dirichlet Process Mixture Model (DPMM) clustering.

Main Results:

• * The BERT-based recommender demonstrated superior performance compared to baseline methods (BM25, TF-IDF, doc2vec, Naïve Bayes) in both internal and external evaluations.
• * Key performance metrics including Recall@k, Precision@k, Mean Reciprocal Rank (MRR), and Area Under the ROC Curve (ROC-AUC) favored the proposed system.
• * External evaluation confirmed the system's practical utility by achieving high Precision@k based on researcher ratings.

Conclusions:

• * The proposed BERT-based recommendation system effectively addresses the challenge of identifying relevant NIH grants for researchers.
• * Deep learning models like BERT can capture nuanced connections between research output and funding opportunities.
• * This system offers a promising solution to enhance grant discovery efficiency and support researchers' academic endeavors.