پیش‌بینی تأثیرگذاری پژوهش‌های علمی با استفاده از الگوریتم‌های یادگیری ماشین

Abstract
Purpose: "Research impact" is one of the key concerns of the stakeholders of this field, which can be considered as positive and profitable applications of research in various dimensions such as society, economy, environment, culture, health, etc. The current study aims to measure the relationship of each of the different variables of scientific outputs impacts on each other and also to exam which of the machine algorithms can predict the scientific, social and economic impact of scientific outputs.
Methodology: The current research is applied in terms of purpose and descriptive in terms of method and has been done by scientometric approach. This study aims to investigate the relationship between the volume of scientific outputs and scientific cooperation as well as the scientific, social and economic impact of Iran's scientific outputs in biotechnology; and which of the machine learning algorithms are better predictors for measuring the effectiveness of scientific outputs in different dimensions. The research community is Iran's biotechnology scientific outputs, which are indexed in the Scopus database in the period of 2003-2024. The data extracted 1 Bahman 1402 equivalent to 21 January 2024. In order to extract relevant data, Scival analytical database was used. In this research, Pearson's correlation coefficient and R software package were used to determine the relationship between the studied indicators. Machine learning algorithms such as multiple linear regression, nearest neighbor, decision trees, random forests and gradient boosting were also used and evaluated as predictive models. In order to perform tests and algorithms, Python programming language has been used.
Findings: The findings of this study showed that Iran's scientific outputs in this field in the period from 2003 to 2023 had an increased 36 times, which is considered an extremely high progress. The relationship between international collaboration and indicators such as citations counts, Field-Weighted Citation Impact, Output in Top 10% Citation Percentiles, Patent-Citations Count, Patent-Citations per Scholarly Output, Scholarly Output cited by Patents, Patents Count, Views Count, Output in Top 10% Views Percentiles, Views per Publication, and Field-Weighted Views Impact of the domain were found to be positive and significant. The index of academic collaboration also has a positive and significant relationship with the citations counts, Field-Weighted Citation Impact, Output in Top 10% Citation Percentiles, Publications in Top 10% Journal Percentiles by Cite Score Percentile, Patents Count, the Scholarly Output cited by Patents, Patents Count, Views Count and Field-Weighted Views Impact. Academic-Government Collaboration also has a positive and significant relationship with three indicators citations per publication, Patent-Citations Count and Patent-Citations per Scholarly Output. In the case of impact of the studied scientific outputs, citations counts has positive and significant relationship with the indicators Scholarly Output cited by Patents, Patents Count, Views Count, Views per Publication and the Field-Weighted Views Impact of the biotechnology scientific outputs. In economic impact, the result indicated that the number of patent citations as one of the representative indicators of this type of impact, with indicators such as academic Collaboration, international Collaboration, citations counts, citations per Publications, Field-Weighted Citation Impact, Views Count has positive and significant relationship in Output in Top 10% Views Percentiles, Views per Publication and Field-Weighted Views Impact the biotechnology scientific outputs. In terms of social impact, it was also concluded that Views Count has positive and significant relationship with indicators such as citations counts, Field-Weighted Citation Impact, the number of patent citations, Patent-Citations per Scholarly Output, Scholarly Output cited by Patents and Patents Count in biotechnology field. Based on the obtained results, multivariate linear regression with a higher accuracy score and a lower standard deviation score could better predict the scientific, technological and social impact of Iran's scientific outputs in biotechnology.
Conclusion: The most important factor affecting the quality of articles, including citations, views, and applicability, is international cooperation, and it is necessary to about measures in this regard. It is suggested to use a variety of indicators during the quantitative and qualitative evaluation of the articles in order to obtain a clearer picture of the effectiveness of the research. The important point for policymakers in science and technology for the scientific development in Iran is that despite the quantitative and qualitative growth of scientific outputs, the country's research system should be directed towards value creation and creating added value, especially in the economic field.