# Whats next? Forecasting scientific research trends

**Authors:** Dan Ofer, Michal Linial  
**arXiv:** [2305.04133](https://arxiv.org/abs/2305.04133)  
**Date:** 2023-05-06  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.04133v1/

## Key points

- Scientific research trends and interests are constantly evolving
- It is crucial for educational institutions, practitioners, investors, and funding organizations to identify and forecast these trends
- Researchers aimed to predict future trends in scientific publications using a diverse list of 127 topics focused mostly on life science
- Heterogeneous public sources such as historical publications from PubMed, research and review articles, and patents were used to train forecasting models that could predict topic popularity five years in advance
- Normalized fraction of citations a paper received out of the total number of citations in PubMed each year was used instead of absolute quantities like the number of citations in a year
- Preceding publications and future patents served as leading indicators for emerging scientific topics
- Declining topics had an excess of reviews compared to original research articles while identifying increasing or declining topics based on their ratio
- Language models were used to provide improved insights into topic temporal dynamics across molecular, technological, and conceptual areas within biomedical research.
- The study provides a valuable framework for identifying and predicting scientific research trends using heterogeneous public sources.
- Long-term forecasting in this domain is possible.

## Summary

Scientific research trends and interests are constantly evolving, making it crucial for educational institutions, practitioners, investors, and funding organizations to identify and forecast these trends. In this study, the researchers aimed to predict future trends in scientific publications using a diverse list of 127 topics focused mostly on life science. These topics included neuroanatomical regions, experimental methods, emerging biomedical technology, domain-expert selections for science, technology, and engineering research topics. The team used heterogeneous public sources such as historical publications from PubMed, research and review articles, and patents to train forecasting models that could predict topic popularity five years in advance. Instead of absolute quantities like the number of citations in a year, they used the normalized fraction of citations a paper received out of the total number of citations in PubMed each year. They also calculated an absolute quantity of citations per topic by multiplying the normalized frequency by the number of Medline publications that year. The researchers found that preceding publications and future patents served as leading indicators for emerging scientific topics. They also discovered that declining topics had an excess of reviews compared to original research articles while identifying increasing or declining topics based on their ratio. The team used language models to provide improved insights into topic temporal dynamics across molecular, technological, and conceptual areas within biomedical research. While bibliometric publication trend statistics are not a perfect approximation for scientific research's direction entirely; they can provide valuable insights into trends within a field. The researchers demonstrated that it is possible to identify emerging trends in a field and predict how they may evolve over time with some degree of accuracy. By training forecasting models capable of predicting future topic popularity five years ahead better than naïve baselines; the researchers showed that long-term forecasting in this domain is possible. In conclusion; this study provides a valuable framework for identifying and predicting scientific research trends using heterogeneous public sources. The findings suggest that similar dynamics apply to molecular; technological; and conceptual topics across biomedical research. This approach can help educational institutions; practitioners; investors; and funding organizations make informed decisions about future research directions.